Intel Arc Pro B70 vs AMD EPYC 7002 / 7003 series NAS (512 GB) for local AI

Pick two to four systems or devices to see which runs more models, faster, for the kind of work you do.

Hardware to compare, two to four

Winner: AMD EPYC 7002 / 7003 series NAS (512 GB)

For general chat: It runs 101 more LLMs (614 vs 513). On the other hand, the Intel Arc Pro B70 is about 12.4× faster on its own top models.

Side by side

Swap sides All models on the Arc Pro B70 All models on the EPYC 7002 / 7003 series NAS

  Intel Arc Pro B70Intel Arc Pro B70 QNAP QAI-h1290FXAMD EPYC 7002 / 7003 series NAS (512 GB)
Memory 32.0 GB VRAM 512 GB unified
Usable for models 32.0 GB 471 GB
Memory bandwidth 608 GB/s 205 GB/s
FP16 compute 37.4 TFLOPS 1.0 TFLOPS
Launch price $949 —
Released 2026 2019
LLMs that run 513 of 619 614 of 619
LLMs that run well 484 459
Image models that run 59 of 60 60 of 60
Video models that run 40 of 43 43 of 43
Decision models that run 68 of 70 70 of 70
Top-10 LLM average score 100 91
Top-10 LLM average speed 84.3 tok/s 6.8 tok/s
Best LLM MiniCPM-V 4.7 35B-A3B MiniMax M3.1 Flash (428B-A23B)

Highlighted values are the best of the two. Model results assume 8K of context and the default macOS GPU memory limit on Macs.

LLMs

513 of 619 LLMs run on the Intel Arc Pro B70. 32.0 GB usable for models.

614 of 619 LLMs run on the AMD EPYC 7002 / 7003 series NAS (512 GB). 471 GB usable for models.

Model QualityArc Pro B70 SpeedArc Pro B70 RunsArc Pro B70 QualityEPYC 7002 / 7003 series NAS SpeedEPYC 7002 / 7003 series NAS RunsEPYC 7002 / 7003 series NAS
MiniCPM-V 4.7 35B-A3BOpenBMB · 35B (3.0B active) 92Excellent quality (A+) · 92 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 100/100 (quality, fit, speed and memory use) sets the ranking. 89.3 tok/s Runs welltight fit 88Excellent quality (A+) · 88 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 87/100 (quality, fit, speed and memory use) sets the ranking. 40.2 tok/s Runs great
Qwen3.5 35B-A3BAlibaba · 36B (3.0B active) 92Excellent quality (A+) · 92 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 100/100 (quality, fit, speed and memory use) sets the ranking. 89.3 tok/s Runs welltight fit 88Excellent quality (A+) · 88 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 87/100 (quality, fit, speed and memory use) sets the ranking. 40.2 tok/s Runs great
Qwen3.6 35B-A3BAlibaba · 36B (3.0B active) 92Excellent quality (A+) · 92 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 100/100 (quality, fit, speed and memory use) sets the ranking. 89.3 tok/s Runs welltight fit 88Excellent quality (A+) · 88 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 87/100 (quality, fit, speed and memory use) sets the ranking. 40.2 tok/s Runs great
Gemma 4 26B-A4BGoogle · 26B (3.8B active) 92Excellent quality (A+) · 92 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 100/100 (quality, fit, speed and memory use) sets the ranking. 59.2 tok/s Runs welltight fit 87Excellent quality (A+) · 87 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 85/100 (quality, fit, speed and memory use) sets the ranking. 32.7 tok/s Runs great
Nex N2 Mini (35B-A3B)Nex-AGI · 35B (3.0B active) 91Excellent quality (A+) · 91 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 99/100 (quality, fit, speed and memory use) sets the ranking. 89.3 tok/s Runs welltight fit 87Excellent quality (A+) · 87 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 87/100 (quality, fit, speed and memory use) sets the ranking. 40.2 tok/s Runs great
Sage-A1.5 35B-A3BInternScience · 36B (3.0B active) 90Excellent quality (A+) · 90 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 99/100 (quality, fit, speed and memory use) sets the ranking. 89.3 tok/s Runs welltight fit 86Excellent quality (A+) · 86 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 86/100 (quality, fit, speed and memory use) sets the ranking. 40.2 tok/s Runs great
Raon-1 mini Preview (30B-A4B)KRAFTON · 29B (4.3B active) 87Excellent quality (A+) · 87 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 99/100 (quality, fit, speed and memory use) sets the ranking. 69.1 tok/s Runs great 83Great quality (A) · 83 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 83/100 (quality, fit, speed and memory use) sets the ranking. 29.3 tok/s Runs well
Ornith 1.5 35B-A3BDeepReinforce · 36B (3.0B active) 89Excellent quality (A+) · 89 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 99/100 (quality, fit, speed and memory use) sets the ranking. 89.3 tok/s Runs welltight fit 85Excellent quality (A+) · 85 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 86/100 (quality, fit, speed and memory use) sets the ranking. 40.2 tok/s Runs great
Apodex 1.1 Mini (35B-A3B)Apodex · 36B (3.0B active) 88Excellent quality (A+) · 88 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 98/100 (quality, fit, speed and memory use) sets the ranking. 89.3 tok/s Runs welltight fit 84Great quality (A) · 84 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 86/100 (quality, fit, speed and memory use) sets the ranking. 40.2 tok/s Runs great
Ornith 1.0 35B-A3BDeepReinforce · 36B (3.0B active) 88Excellent quality (A+) · 88 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 98/100 (quality, fit, speed and memory use) sets the ranking. 89.3 tok/s Runs welltight fit 84Great quality (A) · 84 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 86/100 (quality, fit, speed and memory use) sets the ranking. 40.2 tok/s Runs great
Flint (40B-A2.5B)Recursive · 40B (2.5B active) 85Excellent quality (A+) · 85 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 98/100 (quality, fit, speed and memory use) sets the ranking. 129 tok/s Runs great 87Excellent quality (A+) · 87 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 87/100 (quality, fit, speed and memory use) sets the ranking. 39.7 tok/s Runs great
GLM-4.7 Flash (30B-A3B)Zhipu AI · 31B (3.0B active) 85Excellent quality (A+) · 85 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 98/100 (quality, fit, speed and memory use) sets the ranking. 101 tok/s Runs great 83Great quality (A) · 83 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 85/100 (quality, fit, speed and memory use) sets the ranking. 40.2 tok/s Runs great
Qwen AgentWorld 35B-A3BAlibaba · 35B (3.0B active) 84Great quality (A) · 84 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 97/100 (quality, fit, speed and memory use) sets the ranking. 101 tok/s Runs great 82Great quality (A) · 82 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 85/100 (quality, fit, speed and memory use) sets the ranking. 40.2 tok/s Runs great
Agents-A1 35B-A3BInternScience · 35B (3.0B active) 86Excellent quality (A+) · 86 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 97/100 (quality, fit, speed and memory use) sets the ranking. 89.3 tok/s Runs welltight fit 82Great quality (A) · 82 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 85/100 (quality, fit, speed and memory use) sets the ranking. 40.2 tok/s Runs great
Qwen3.8 27BAlibaba · 28B 90Excellent quality (A+) · 90 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 95/100 (quality, fit, speed and memory use) sets the ranking. 20.5 tok/s Runs well 94Excellent quality (A+) · 94 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. 3.5 tok/s Slow
Nanbeige4.3 33BBOSS Zhipin (Nanbeige) · 33B 90Excellent quality (A+) · 90 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 95/100 (quality, fit, speed and memory use) sets the ranking. 17.2 tok/s Runs well 92Excellent quality (A+) · 92 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. 3.5 tok/s Slow
Laguna XS 2.1 (33B-A3B)Poolside · 33B (3.0B active) 81Great quality (A) · 81 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 95/100 (quality, fit, speed and memory use) sets the ranking. 89.3 tok/s Runs welltight fit 77Great quality (A) · 77 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 82/100 (quality, fit, speed and memory use) sets the ranking. 40.2 tok/s Runs great
Qwen3.6 27BAlibaba · 28B 89Excellent quality (A+) · 89 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 95/100 (quality, fit, speed and memory use) sets the ranking. 20.5 tok/s Runs well 93Excellent quality (A+) · 93 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. 3.5 tok/s Slow
Nemotron Cascade 2 30B-A3BNVIDIA · 32B (3.5B active) 77Great quality (A) · 77 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 94/100 (quality, fit, speed and memory use) sets the ranking. 91.5 tok/s Runs great 76Great quality (A) · 76 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking. 35.2 tok/s Runs great
DiffusionGemma 26B-A4B ITGoogle · 26B (4.0B active) 77Great quality (A) · 77 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 94/100 (quality, fit, speed and memory use) sets the ranking. 72.9 tok/s Runs great 74Great quality (A) · 74 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. 31.2 tok/s Runs great
KAT-Coder V2.5 Dev (35B-A3B)Kuaishou · 35B (3.0B active) 88Excellent quality (A+) · 88 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 94/100 (quality, fit, speed and memory use) sets the ranking. 101 tok/s Runs great 86Excellent quality (A+) · 86 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking. 40.2 tok/s Runs great
Qwen3.5 27BAlibaba · 28B 88Excellent quality (A+) · 88 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 94/100 (quality, fit, speed and memory use) sets the ranking. 20.5 tok/s Runs well 92Excellent quality (A+) · 92 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 78/100 (quality, fit, speed and memory use) sets the ranking. 3.5 tok/s Slow
Muse Glimmer 30BMeta · 30B 87Excellent quality (A+) · 87 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 94/100 (quality, fit, speed and memory use) sets the ranking. 19.1 tok/s Runs well 91Excellent quality (A+) · 91 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 78/100 (quality, fit, speed and memory use) sets the ranking. 3.3 tok/s Slow
Gemma 4 31BGoogle · 31B 86Excellent quality (A+) · 86 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 94/100 (quality, fit, speed and memory use) sets the ranking. 18.2 tok/s Runs well 90Excellent quality (A+) · 90 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking. 3.1 tok/s Slow
Nemotron 3 Nano 30B-A3BNVIDIA · 32B (3.5B active) 74Great quality (A) · 74 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 93/100 (quality, fit, speed and memory use) sets the ranking. 91.5 tok/s Runs great 73Great quality (A) · 73 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. 35.2 tok/s Runs great
Qwen3 30B-A3B Instruct 2507Alibaba · 31B (3.3B active) 74Great quality (A) · 74 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 93/100 (quality, fit, speed and memory use) sets the ranking. 95.2 tok/s Runs great 73Great quality (A) · 73 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. 37.0 tok/s Runs great
TMax 27BAi2 · 27B 85Excellent quality (A+) · 85 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 93/100 (quality, fit, speed and memory use) sets the ranking. 20.8 tok/s Runs well 89Excellent quality (A+) · 89 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking. 3.5 tok/s Slow
Qwen3-VL 30B-A3B ThinkingAlibaba · 31B (3.3B active) 73Great quality (A) · 73 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 93/100 (quality, fit, speed and memory use) sets the ranking. 95.2 tok/s Runs great 72Great quality (A) · 72 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. 37.0 tok/s Runs great
LFM2.5 24B-A2BLiquid AI · 24B (2.5B active) 73Great quality (A) · 73 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 92/100 (quality, fit, speed and memory use) sets the ranking. 102 tok/s Runs great 71Great quality (A) · 71 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. 40.4 tok/s Runs great
Fara 1.5 27BMicrosoft · 27B 83Great quality (A) · 83 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 92/100 (quality, fit, speed and memory use) sets the ranking. 20.8 tok/s Runs well 87Excellent quality (A+) · 87 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking. 3.5 tok/s Slow
Qwen3 30B-A3BAlibaba · 31B (3.3B active) 71Great quality (A) · 71 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 92/100 (quality, fit, speed and memory use) sets the ranking. 95.2 tok/s Runs great 70Great quality (A) · 70 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 78/100 (quality, fit, speed and memory use) sets the ranking. 37.0 tok/s Runs great
Qwen3-VL 30B-A3B InstructAlibaba · 31B (3.3B active) 71Great quality (A) · 71 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 92/100 (quality, fit, speed and memory use) sets the ranking. 95.2 tok/s Runs great 70Great quality (A) · 70 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 78/100 (quality, fit, speed and memory use) sets the ranking. 37.0 tok/s Runs great
Hunyuan Hy-Memory 30B-A3BTencent · 30B (3.0B active) 82Great quality (A) · 82 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 92/100 (quality, fit, speed and memory use) sets the ranking. 101 tok/s Runs great 80Great quality (A) · 80 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. 40.2 tok/s Runs great
UI-Mate 27BTencent · 27B 81Great quality (A) · 81 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 91/100 (quality, fit, speed and memory use) sets the ranking. 20.8 tok/s Runs well 85Excellent quality (A+) · 85 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking. 3.5 tok/s Slow
Nemotron 3.5 Lightning 30B-A3BNVIDIA · 32B (3.0B active) 69Good quality (B) · 69 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 91/100 (quality, fit, speed and memory use) sets the ranking. 101 tok/s Runs great 68Good quality (B) · 68 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 78/100 (quality, fit, speed and memory use) sets the ranking. 40.2 tok/s Runs great
Sarvam 30BSarvam AI · 32B (2.4B active) 67Good quality (B) · 67 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 90/100 (quality, fit, speed and memory use) sets the ranking. 116 tok/s Runs great 67Good quality (B) · 67 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 78/100 (quality, fit, speed and memory use) sets the ranking. 41.2 tok/s Runs great
Qwen3-Omni 30B-A3B InstructAlibaba · 35B (3.3B active) 69Good quality (B) · 69 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 90/100 (quality, fit, speed and memory use) sets the ranking. 83.7 tok/s Runs welltight fit 66Good quality (B) · 66 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking. 37.0 tok/s Runs great
North Mini Code 1.0 (30B-A3B)Cohere · 30B (3.0B active) 77Great quality (A) · 77 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 89/100 (quality, fit, speed and memory use) sets the ranking. 101 tok/s Runs great 76Great quality (A) · 76 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking. 40.2 tok/s Runs great
EXAONE 4.5 33BLG AI Research · 34B 77Great quality (A) · 77 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 89/100 (quality, fit, speed and memory use) sets the ranking. 16.6 tok/s Runs well 79Great quality (A) · 79 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. 3.4 tok/s Slow
Trinity Mini (26B-A3B)Arcee AI · 26B (3.0B active) 65Good quality (B) · 65 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 89/100 (quality, fit, speed and memory use) sets the ranking. 89.3 tok/s Runs great 62Good quality (B) · 62 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking. 40.2 tok/s Runs great
Gemma 4 12BGoogle · 12B 80Great quality (A) · 80 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 89/100 (quality, fit, speed and memory use) sets the ranking. 39.4 tok/s Runs great 78Great quality (A) · 78 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking. 11.7 tok/s Runs well
Qwen3 30B-A3B Thinking 2507Alibaba · 31B (3.3B active) 75Great quality (A) · 75 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 89/100 (quality, fit, speed and memory use) sets the ranking. 95.2 tok/s Runs great 74Great quality (A) · 74 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking. 37.0 tok/s Runs great
Qwen3 Coder 30B-A3BAlibaba · 31B (3.3B active) 75Great quality (A) · 75 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 89/100 (quality, fit, speed and memory use) sets the ranking. 95.2 tok/s Runs great 74Great quality (A) · 74 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking. 37.0 tok/s Runs great
ERNIE 4.5 21B-A3BBaidu · 22B (3.0B active) 64Good quality (B) · 64 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 88/100 (quality, fit, speed and memory use) sets the ranking. 70.7 tok/s Runs great 60Good quality (B) · 60 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. 40.2 tok/s Runs great
Seed-OSS 36B InstructByteDance · 36B 76Great quality (A) · 76 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 88/100 (quality, fit, speed and memory use) sets the ranking. 15.8 tok/s Runs well 77Great quality (A) · 77 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. 3.2 tok/s Slow
K2 Horizon 7BInstitute of Foundation Models · 9.0B 79Great quality (A) · 79 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 88/100 (quality, fit, speed and memory use) sets the ranking. 44.0 tok/s Runs great 76Great quality (A) · 76 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking. 15.5 tok/s Runs well
Hermes 4.3 36BNous Research · 36B 75Great quality (A) · 75 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 88/100 (quality, fit, speed and memory use) sets the ranking. 15.8 tok/s Runs well 76Great quality (A) · 76 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. 3.2 tok/s Slow
Qwen3-VL 32B InstructAlibaba · 33B 74Great quality (A) · 74 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 88/100 (quality, fit, speed and memory use) sets the ranking. 17.1 tok/s Runs well 75Great quality (A) · 75 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. 3.5 tok/s Slow
Kimi Linear 48B-A3BMoonshot AI · 49B (3.0B active) 62Good quality (B) · 62 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 87/100 (quality, fit, speed and memory use) sets the ranking. 137 tok/s Runs great 70Great quality (A) · 70 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. 40.2 tok/s Runs great
Granite 4.2 30BIBM · 29B 72Great quality (A) · 72 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 87/100 (quality, fit, speed and memory use) sets the ranking. 19.5 tok/s Runs well 75Great quality (A) · 75 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. 3.3 tok/s Slow
SenseNova Loop 9B PreviewSenseTime · 9.7B 77Great quality (A) · 77 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 87/100 (quality, fit, speed and memory use) sets the ranking. 40.9 tok/s Runs great 74Great quality (A) · 74 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. 14.5 tok/s Runs well
Qwen3 32BAlibaba · 33B 72Great quality (A) · 72 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 87/100 (quality, fit, speed and memory use) sets the ranking. 17.4 tok/s Runs well 73Great quality (A) · 73 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking. 3.5 tok/s Slow
Devstral Small 2 (24B)Mistral AI · 24B 72Great quality (A) · 72 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 87/100 (quality, fit, speed and memory use) sets the ranking. 23.7 tok/s Runs well 75Great quality (A) · 75 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. 4.0 tok/s Slow
Granite 4.0 H Small (32B-A9B)IBM · 32B (9.0B active) 60Good quality (B) · 60 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 87/100 (quality, fit, speed and memory use) sets the ranking. 44.2 tok/s Runs great 58Good quality (B) · 58 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 68/100 (quality, fit, speed and memory use) sets the ranking. 14.7 tok/s Runs well
gpt-oss 20BOpenAI · 21B (3.6B active) 70Great quality (A) · 70 of 100: how capable the model is at MXFP4, whatever it runs on. Match score 86/100 (quality, fit, speed and memory use) sets the ranking. 108 tok/s Runs great 70Great quality (A) · 70 of 100: how capable the model is at MXFP4, whatever it runs on. Match score 78/100 (quality, fit, speed and memory use) sets the ranking. 36.4 tok/s Runs great
Ornith 1.5 9BDeepReinforce · 9.7B 76Great quality (A) · 76 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 86/100 (quality, fit, speed and memory use) sets the ranking. 41.0 tok/s Runs great 73Great quality (A) · 73 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. 14.5 tok/s Runs well
Axion 1.0 35BAxion · 35B (3.0B active) 61Good quality (B) · 61 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 86/100 (quality, fit, speed and memory use) sets the ranking. 89.3 tok/s Runs welltight fit 58Good quality (B) · 58 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. 40.2 tok/s Runs great
GLM-4 32B 0414Zhipu AI · 33B 70Great quality (A) · 70 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 86/100 (quality, fit, speed and memory use) sets the ranking. 17.5 tok/s Runs well 71Great quality (A) · 71 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 3.5 tok/s Slow
ERNIE 4.5 21B-A3B ThinkingBaidu · 22B (3.0B active) 70Great quality (A) · 70 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 86/100 (quality, fit, speed and memory use) sets the ranking. 70.7 tok/s Runs great 66Good quality (B) · 66 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. 40.2 tok/s Runs great
Maple Preview (20B-A1B)DeepGrove · 20B (1.5B active) 70Great quality (A) · 70 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 86/100 (quality, fit, speed and memory use) sets the ranking. 112 tok/s Runs great 70Great quality (A) · 70 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. 40.4 tok/s Runs great
Mellum2.1 Thinking (12B-A2.5B)JetBrains · 12B (2.5B active) 70Great quality (A) · 70 of 100: how capable the model is at F16, whatever it runs on. Match score 86/100 (quality, fit, speed and memory use) sets the ranking. 43.4 tok/s Runs great 67Good quality (B) · 67 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. 39.7 tok/s Runs great
Qwen3.5 9BAlibaba · 9.7B 75Great quality (A) · 75 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 86/100 (quality, fit, speed and memory use) sets the ranking. 41.0 tok/s Runs great 72Great quality (A) · 72 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. 14.5 tok/s Runs well
Granite 4.1 30BIBM · 29B 68Good quality (B) · 68 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 86/100 (quality, fit, speed and memory use) sets the ranking. 19.7 tok/s Runs well 71Great quality (A) · 71 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 3.4 tok/s Slow
Kimi-VL A3B Thinking 2506Moonshot AI · 16B (3.0B active) 60Good quality (B) · 60 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 86/100 (quality, fit, speed and memory use) sets the ranking. 70.7 tok/s Runs great 57Good quality (B) · 57 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. 40.2 tok/s Runs great
Ornith 1.0 9BDeepReinforce · 9.7B 74Great quality (A) · 74 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 85/100 (quality, fit, speed and memory use) sets the ranking. 41.0 tok/s Runs great 71Great quality (A) · 71 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. 14.5 tok/s Runs well
Gemma 3 27B ITGoogle · 27B 68Good quality (B) · 68 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 85/100 (quality, fit, speed and memory use) sets the ranking. 20.8 tok/s Runs well 71Great quality (A) · 71 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 3.5 tok/s Slow
EXAONE 4.0 32BLG AI Research · 32B 68Good quality (B) · 68 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 85/100 (quality, fit, speed and memory use) sets the ranking. 17.8 tok/s Runs well 71Great quality (A) · 71 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 3.0 tok/s Slow
Olmo 3.1 32B InstructAi2 · 32B 68Good quality (B) · 68 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 85/100 (quality, fit, speed and memory use) sets the ranking. 17.7 tok/s Runs well 71Great quality (A) · 71 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 3.0 tok/s Slow
Groot 19B A3B 261023GrootTeam · 19B (3.0B active) 57Good quality (B) · 57 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 85/100 (quality, fit, speed and memory use) sets the ranking. 70.7 tok/s Runs great 54Fair quality (C) · 54 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. 40.2 tok/s Runs great
InternVL3.5 38BShanghai AI Lab · 38B 72Great quality (A) · 72 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 85/100 (quality, fit, speed and memory use) sets the ranking. 14.9 tok/s Runs welltight fit 73Great quality (A) · 73 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking. 3.0 tok/s Slow
NorthCUA Minibeipo-ai · 35B (3.0B active) 58Good quality (B) · 58 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 85/100 (quality, fit, speed and memory use) sets the ranking. 89.3 tok/s Runs welltight fit 56Good quality (B) · 56 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. 40.2 tok/s Runs great
Ling 3.0 Tiny (8B-A1.3B)Ant Group (inclusionAI) · 7.9B (1.3B active) 60Good quality (B) · 60 of 100: how capable the model is at F16, whatever it runs on. Match score 85/100 (quality, fit, speed and memory use) sets the ranking. 71.0 tok/s Runs great 60Good quality (B) · 60 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. 45.3 tok/s Runs great
TMax 9BAi2 · 9.0B 73Great quality (A) · 73 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 85/100 (quality, fit, speed and memory use) sets the ranking. 44.2 tok/s Runs great 70Great quality (A) · 70 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. 15.6 tok/s Runs well
LFM2 24B-A2BLiquid AI · 24B (2.3B active) 55Good quality (B) · 55 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 84/100 (quality, fit, speed and memory use) sets the ranking. 106 tok/s Runs great 54Fair quality (C) · 54 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. 42.7 tok/s Runs great
MedGemma 27B Text ITGoogle · 27B 66Good quality (B) · 66 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 84/100 (quality, fit, speed and memory use) sets the ranking. 21.1 tok/s Runs well 69Good quality (B) · 69 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 68/100 (quality, fit, speed and memory use) sets the ranking. 3.6 tok/s Slow
Baichuan M2 32BBaichuan · 33B 66Good quality (B) · 66 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 84/100 (quality, fit, speed and memory use) sets the ranking. 17.4 tok/s Runs well 67Good quality (B) · 67 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 68/100 (quality, fit, speed and memory use) sets the ranking. 3.5 tok/s Slow
Qwen2.5 32B InstructAlibaba · 33B 66Good quality (B) · 66 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 84/100 (quality, fit, speed and memory use) sets the ranking. 17.4 tok/s Runs well 67Good quality (B) · 67 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 68/100 (quality, fit, speed and memory use) sets the ranking. 3.5 tok/s Slow
Apriel 1.5 15B ThinkerServiceNow · 15B 68Good quality (B) · 68 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 84/100 (quality, fit, speed and memory use) sets the ranking. 38.4 tok/s Runs great 60Good quality (B) · 60 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 12.2 tok/s Runs well
Devstral Small 1.1 (24B)Mistral AI · 24B 66Good quality (B) · 66 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 84/100 (quality, fit, speed and memory use) sets the ranking. 24.2 tok/s Runs well 69Good quality (B) · 69 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 68/100 (quality, fit, speed and memory use) sets the ranking. 4.1 tok/s Slow
Magistral Small 2507Mistral AI · 24B 66Good quality (B) · 66 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 84/100 (quality, fit, speed and memory use) sets the ranking. 24.2 tok/s Runs well 69Good quality (B) · 69 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 68/100 (quality, fit, speed and memory use) sets the ranking. 4.1 tok/s Slow
Nemotron 3.5 Bolt E4BNVIDIA · 8.0B 70Great quality (A) · 70 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 84/100 (quality, fit, speed and memory use) sets the ranking. 64.5 tok/s Runs great 66Good quality (B) · 66 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking. 31.0 tok/s Runs great
Terminal RSI 27BTeriminal-RSI · 28B 64Good quality (B) · 64 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 84/100 (quality, fit, speed and memory use) sets the ranking. 20.5 tok/s Runs well 67Good quality (B) · 67 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 3.5 tok/s Slow
Phi-4 Reasoning Vision 15BMicrosoft · 15B 66Good quality (B) · 66 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 83/100 (quality, fit, speed and memory use) sets the ranking. 37.7 tok/s Runs great 58Good quality (B) · 58 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 11.9 tok/s Runs well
Qwen2.5-VL 32B InstructAlibaba · 33B 64Good quality (B) · 64 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 83/100 (quality, fit, speed and memory use) sets the ranking. 17.0 tok/s Runs well 65Good quality (B) · 65 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 3.4 tok/s Slow
Mistral Small 3.2 24BMistral AI · 24B 64Good quality (B) · 64 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 83/100 (quality, fit, speed and memory use) sets the ranking. 23.7 tok/s Runs well 67Good quality (B) · 67 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 4.0 tok/s Slow
Magistral Small 2506Mistral AI · 24B 64Good quality (B) · 64 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 83/100 (quality, fit, speed and memory use) sets the ranking. 24.2 tok/s Runs well 67Good quality (B) · 67 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 4.1 tok/s Slow
Gemma 4 E4BGoogle · 8.0B 68Good quality (B) · 68 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 83/100 (quality, fit, speed and memory use) sets the ranking. 64.5 tok/s Runs great 64Good quality (B) · 64 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. 31.0 tok/s Runs great
Hemmingway 1.5Hemmingway · 27B 63Good quality (B) · 63 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 83/100 (quality, fit, speed and memory use) sets the ranking. 21.2 tok/s Runs well 66Good quality (B) · 66 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 3.6 tok/s Slow
Llama 3.3 Nemotron Super 49B v1.5NVIDIA · 50B 67Good quality (B) · 67 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 83/100 (quality, fit, speed and memory use) sets the ranking. 14.8 tok/s Runs welltight fit 67Good quality (B) · 67 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 3.6 tok/s Slow
QwQ 32BAlibaba · 33B 74Great quality (A) · 74 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 83/100 (quality, fit, speed and memory use) sets the ranking. 17.4 tok/s Runs well 75Great quality (A) · 75 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 3.5 tok/s Slow
InternVL3 38BShanghai AI Lab · 38B 66Good quality (B) · 66 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 83/100 (quality, fit, speed and memory use) sets the ranking. 14.9 tok/s Runs welltight fit 67Good quality (B) · 67 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 3.0 tok/s Slow
Granite 4.2 8BIBM · 8.8B 67Good quality (B) · 67 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 83/100 (quality, fit, speed and memory use) sets the ranking. 45.0 tok/s Runs great 64Good quality (B) · 64 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking. 15.9 tok/s Runs well
Phi-3.5 MoE (42B-A6.6B)Microsoft · 42B (6.6B active) 53Fair quality (C) · 53 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 82/100 (quality, fit, speed and memory use) sets the ranking. 66.9 tok/s Runs welltight fit 53Fair quality (C) · 53 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 19.8 tok/s Runs well
Mistral Small 3.1 24BMistral AI · 24B 62Good quality (B) · 62 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 82/100 (quality, fit, speed and memory use) sets the ranking. 23.8 tok/s Runs well 65Good quality (B) · 65 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 4.0 tok/s Slow
Qwen3 14BAlibaba · 15B 64Good quality (B) · 64 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 82/100 (quality, fit, speed and memory use) sets the ranking. 38.5 tok/s Runs great 57Good quality (B) · 57 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 12.2 tok/s Runs well
Ling Mini 2.0 (16B-A1.4B)Ant Group (inclusionAI) · 16B (1.4B active) 52Fair quality (C) · 52 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 82/100 (quality, fit, speed and memory use) sets the ranking. 116 tok/s Runs great 52Fair quality (C) · 52 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. 42.6 tok/s Runs great
Olmo 3.1 32B ThinkAi2 · 32B 72Great quality (A) · 72 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 82/100 (quality, fit, speed and memory use) sets the ranking. 17.7 tok/s Runs well 75Great quality (A) · 75 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 3.0 tok/s Slow
OpenReasoning Nemotron 32BNVIDIA · 33B 72Great quality (A) · 72 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 82/100 (quality, fit, speed and memory use) sets the ranking. 17.4 tok/s Runs well 73Great quality (A) · 73 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 3.5 tok/s Slow
Skywork OR1 32BSkywork · 33B 72Great quality (A) · 72 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 82/100 (quality, fit, speed and memory use) sets the ranking. 17.4 tok/s Runs well 73Great quality (A) · 73 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 3.5 tok/s Slow
LFM2.5 8B-A1BLiquid AI · 8.5B (1.5B active) 52Fair quality (C) · 52 of 100: how capable the model is at F16, whatever it runs on. Match score 82/100 (quality, fit, speed and memory use) sets the ranking. 64.1 tok/s Runs great 52Fair quality (C) · 52 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking. 40.2 tok/s Runs great
Gemma 2 27B ITGoogle · 27B 58Good quality (B) · 58 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 82/100 (quality, fit, speed and memory use) sets the ranking. 21.0 tok/s Runs well 61Good quality (B) · 61 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 3.6 tok/s Slow
Falcon-H1 34B InstructTII · 34B 60Good quality (B) · 60 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 82/100 (quality, fit, speed and memory use) sets the ranking. 16.9 tok/s Runs well 62Good quality (B) · 62 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 3.4 tok/s Slow
MiMo V2.6 Distill Qwen 9BXiaomi · 9.4B 77Great quality (A) · 77 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 82/100 (quality, fit, speed and memory use) sets the ranking. 42.0 tok/s Runs great 74Great quality (A) · 74 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 14.8 tok/s Runs well
Hermes 4 14BNous Research · 15B 62Good quality (B) · 62 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 82/100 (quality, fit, speed and memory use) sets the ranking. 38.6 tok/s Runs great 55Good quality (B) · 55 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 12.2 tok/s Runs well
Mistral Small 3 24B (2501)Mistral AI · 24B 60Good quality (B) · 60 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking. 24.2 tok/s Runs well 63Good quality (B) · 63 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 4.1 tok/s Slow
Gemma 4 E2BGoogle · 5.1B 62Good quality (B) · 62 of 100: how capable the model is at F16, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking. 58.2 tok/s Runs great 61Good quality (B) · 61 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. 42.1 tok/s Runs great
Kimi-VL A3B InstructMoonshot AI · 16B (3.0B active) 50Fair quality (C) · 50 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking. 70.7 tok/s Runs great 47Fair quality (C) · 47 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 68/100 (quality, fit, speed and memory use) sets the ranking. 40.2 tok/s Runs great
EXAONE Deep 32BLG AI Research · 32B 70Great quality (A) · 70 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking. 17.8 tok/s Runs well 73Great quality (A) · 73 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 3.0 tok/s Slow
GLM-Z1 32B (0414)Zhipu AI · 33B 70Great quality (A) · 70 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking. 17.5 tok/s Runs well 71Great quality (A) · 71 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 3.5 tok/s Slow
Command R (08-2024) 32BCohere · 32B 58Good quality (B) · 58 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking. 17.7 tok/s Runs well 61Good quality (B) · 61 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 3.0 tok/s Slow
Nanbeige4.2 3BBOSS Zhipin (Nanbeige) · 4.2B 68Good quality (B) · 68 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking. 69.6 tok/s Runs great 64Good quality (B) · 64 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. 33.5 tok/s Runs great
Phi-4 (14B)Microsoft · 15B 60Good quality (B) · 60 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking. 38.8 tok/s Runs great 53Fair quality (C) · 53 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 12.3 tok/s Runs well
InternVL3 14BShanghai AI Lab · 15B 60Good quality (B) · 60 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking. 37.7 tok/s Runs great 53Fair quality (C) · 53 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 11.9 tok/s Runs well
Voxtral Small 24BMistral AI · 24B 58Good quality (B) · 58 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking. 23.5 tok/s Runs well 61Good quality (B) · 61 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 4.0 tok/s Slow
Dolphin Mistral 24B Venice EditionCognitive Computations · 24B 58Good quality (B) · 58 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking. 23.7 tok/s Runs well 61Good quality (B) · 61 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 4.0 tok/s Slow
Ministral 3 14BMistral AI · 14B 60Good quality (B) · 60 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking. 40.9 tok/s Runs great 60Good quality (B) · 60 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 10.0 tok/s Runs well
Nous Hermes 2 Mixtral 8x7B DPONous Research · 47B (12.9B active) 49Fair quality (C) · 49 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. 38.5 tok/s Runs welltight fit 49Fair quality (C) · 49 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 10.4 tok/s Runs well
DeepSeek R1 Distill Qwen 32BDeepSeek · 33B 68Good quality (B) · 68 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. 17.4 tok/s Runs well 69Good quality (B) · 69 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 3.5 tok/s Slow
OLMo 2 32B InstructAi2 · 32B 57Good quality (B) · 57 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. 17.7 tok/s Runs well 59Good quality (B) · 59 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 3.0 tok/s Slow
Aya Expanse 32BCohere · 32B 57Good quality (B) · 57 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. 17.7 tok/s Runs well 59Good quality (B) · 59 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 3.0 tok/s Slow
Nemotron Labs Dory 1.0 4BNVIDIA · 4.0B 68Good quality (B) · 68 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. 72.9 tok/s Runs great 64Good quality (B) · 64 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking. 35.1 tok/s Runs great
Qwen2.5 14B InstructAlibaba · 15B 58Good quality (B) · 58 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. 38.8 tok/s Runs great 52Fair quality (C) · 52 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 12.3 tok/s Runs well
Aya Vision 32BCohere · 33B 57Good quality (B) · 57 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. 17.2 tok/s Runs well 58Good quality (B) · 58 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 3.5 tok/s Slow
Magistral Small 1.2 (24B)Mistral AI · 24B 68Good quality (B) · 68 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. 23.7 tok/s Runs well 71Great quality (A) · 71 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 4.0 tok/s Slow
Qwen3-VL 8B InstructAlibaba · 8.8B 61Good quality (B) · 61 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. 45.1 tok/s Runs great 58Good quality (B) · 58 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 68/100 (quality, fit, speed and memory use) sets the ranking. 15.9 tok/s Runs well
Dolphin 3.0 R1 Mistral 24BCognitive Computations · 24B 57Good quality (B) · 57 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. 24.2 tok/s Runs well 59Good quality (B) · 59 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 4.1 tok/s Slow
InternVL3.5 8BShanghai AI Lab · 8.5B 61Good quality (B) · 61 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. 46.4 tok/s Runs great 58Good quality (B) · 58 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 68/100 (quality, fit, speed and memory use) sets the ranking. 16.4 tok/s Runs well
Mixtral 8x7B InstructMistral AI · 47B (12.9B active) 47Fair quality (C) · 47 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. 38.5 tok/s Runs welltight fit 47Fair quality (C) · 47 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. 10.4 tok/s Runs well
Solar Pro Preview 22BUpstage · 22B 57Good quality (B) · 57 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. 25.8 tok/s Runs well 42Fair quality (C) · 42 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. 10.4 tok/s Runs well
OpenHands LM 32BAll Hands AI · 33B 66Good quality (B) · 66 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. 17.4 tok/s Runs well 67Good quality (B) · 67 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 3.5 tok/s Slow
Phi-4 Reasoning Plus (14B)Microsoft · 15B 68Good quality (B) · 68 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. 38.9 tok/s Runs great 60Good quality (B) · 60 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. 12.3 tok/s Runs well
Moonlight 16B-A3B InstructMoonshot AI · 16B (3.0B active) 46Fair quality (C) · 46 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. 70.7 tok/s Runs great 43Fair quality (C) · 43 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 40.2 tok/s Runs great
Spark-X2.6 4BSparkLLM · 4.1B 66Good quality (B) · 66 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. 70.6 tok/s Runs great 62Good quality (B) · 62 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. 34.0 tok/s Runs great
Qwen2.5 14B Instruct 1MAlibaba · 15B 57Good quality (B) · 57 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. 38.6 tok/s Runs great 50Fair quality (C) · 50 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 12.2 tok/s Runs well
Granite 4.1 8BIBM · 8.8B 59Good quality (B) · 59 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. 45.0 tok/s Runs great 57Good quality (B) · 57 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 15.9 tok/s Runs well
Kolibri 1 (78B-A3.5B)Aleph Alpha · 78B (3.5B active) 76Great quality (A) · 76 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. 70.1 tok/s Runs welloffloads to RAM 86Excellent quality (A+) · 86 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 87/100 (quality, fit, speed and memory use) sets the ranking. 35.5 tok/s Runs great
GLM-4.1V 9B ThinkingZhipu AI · 10B 59Good quality (B) · 59 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. 38.4 tok/s Runs great 57Good quality (B) · 57 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 13.6 tok/s Runs well
MiniCPM-V 4.5 8BOpenBMB · 8.7B 59Good quality (B) · 59 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. 45.5 tok/s Runs great 57Good quality (B) · 57 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 16.1 tok/s Runs well
Gemma 3 12B ITGoogle · 12B 58Good quality (B) · 58 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. 38.6 tok/s Runs great 57Good quality (B) · 57 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 11.4 tok/s Runs well
Qwen3 8BAlibaba · 8.2B 59Good quality (B) · 59 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. 48.3 tok/s Runs great 57Good quality (B) · 57 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 17.1 tok/s Runs well
Phi-4 Reasoning (14B)Microsoft · 15B 66Good quality (B) · 66 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 78/100 (quality, fit, speed and memory use) sets the ranking. 38.9 tok/s Runs great 58Good quality (B) · 58 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. 12.3 tok/s Runs well
Nemotron 3.5 Bolt 74B-A9BNVIDIA · 75B (9.0B active) 77Great quality (A) · 77 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 78/100 (quality, fit, speed and memory use) sets the ranking. 34.2 tok/s Runs welloffloads to RAM 91Excellent quality (A+) · 91 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 83/100 (quality, fit, speed and memory use) sets the ranking. 10.4 tok/s Runs well
Nemotron Nano 9B v2NVIDIA · 8.9B 59Good quality (B) · 59 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 78/100 (quality, fit, speed and memory use) sets the ranking. 44.5 tok/s Runs great 57Good quality (B) · 57 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 15.7 tok/s Runs well
Qwen2.5 Coder 32B InstructAlibaba · 33B 63Good quality (B) · 63 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 78/100 (quality, fit, speed and memory use) sets the ranking. 17.4 tok/s Runs well 64Good quality (B) · 64 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 3.5 tok/s Slow
UI-Mate 9BTencent · 9.4B 69Good quality (B) · 69 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 78/100 (quality, fit, speed and memory use) sets the ranking. 42.0 tok/s Runs great 66Good quality (B) · 66 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 14.8 tok/s Runs well
UI-Venus 2 9BAnt Group (inclusionAI) · 9.4B 69Good quality (B) · 69 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 78/100 (quality, fit, speed and memory use) sets the ranking. 42.0 tok/s Runs great 66Good quality (B) · 66 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 14.8 tok/s Runs well
Qwen3.5 4BAlibaba · 4.7B 62Good quality (B) · 62 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 78/100 (quality, fit, speed and memory use) sets the ranking. 62.3 tok/s Runs great 58Good quality (B) · 58 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. 30.0 tok/s Runs well
MiMo-VL 7B RLXiaomi · 8.3B 57Good quality (B) · 57 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 78/100 (quality, fit, speed and memory use) sets the ranking. 47.6 tok/s Runs great 55Good quality (B) · 55 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 16.8 tok/s Runs well
QwQ 32B PreviewAlibaba · 33B 62Good quality (B) · 62 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 78/100 (quality, fit, speed and memory use) sets the ranking. 17.4 tok/s Runs well 64Good quality (B) · 64 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 3.5 tok/s Slow
SenseNova 6.8 Code2VideoSenseTime · 27B 62Good quality (B) · 62 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 78/100 (quality, fit, speed and memory use) sets the ranking. 20.8 tok/s Runs well 65Good quality (B) · 65 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 3.5 tok/s Slow
Jamba 1.5 MiniAI21 Labs · 52B (12.0B active) 42Fair quality (C) · 42 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking. 51.2 tok/s Runs welltight fit 47Fair quality (C) · 47 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. 11.2 tok/s Runs well
Qwen2 57B-A14B InstructAlibaba · 57B (14.0B active) 42Fair quality (C) · 42 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking. 45.0 tok/s Runs welltight fit 42Fair quality (C) · 42 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 12.4 tok/s Runs well
Mistral Small 22B (2409)Mistral AI · 22B 51Fair quality (C) · 51 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking. 25.6 tok/s Runs well 37Basic quality (D) · 37 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 57/100 (quality, fit, speed and memory use) sets the ranking. 10.3 tok/s Runs well
Devstral Small 2505Mistral AI · 24B 62Good quality (B) · 62 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking. 24.2 tok/s Runs well 65Good quality (B) · 65 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. 4.1 tok/s Slow
Ministral 3 8BMistral AI · 8.9B 55Good quality (B) · 55 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking. 44.3 tok/s Runs great 53Fair quality (C) · 53 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 15.7 tok/s Runs well
Qwen3-Next 80B-A3B ThinkingAlibaba · 81B (3.0B active) 72Great quality (A) · 72 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking. 73.8 tok/s Runs welloffloads to RAM 81Great quality (A) · 81 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 86/100 (quality, fit, speed and memory use) sets the ranking. 40.2 tok/s Runs great
MiniCPM5 2BOpenBMB · 2.5B 60Good quality (B) · 60 of 100: how capable the model is at F16, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking. 53.1 tok/s Runs great 59Good quality (B) · 59 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. 38.5 tok/s Runs great
TMax 4BAi2 · 4.2B 61Good quality (B) · 61 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking. 68.9 tok/s Runs great 57Good quality (B) · 57 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. 33.2 tok/s Runs great
Agents-A1 4BInternScience · 4.5B 60Good quality (B) · 60 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking. 63.9 tok/s Runs great 57Good quality (B) · 57 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking. 30.8 tok/s Runs great
Apriel Nemotron 15B ThinkerServiceNow · 15B 62Good quality (B) · 62 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking. 38.1 tok/s Runs great 55Good quality (B) · 55 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 12.1 tok/s Runs well
AceReason Nemotron 14BNVIDIA · 15B 62Good quality (B) · 62 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking. 38.6 tok/s Runs great 55Good quality (B) · 55 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 12.2 tok/s Runs well
OpenReasoning Nemotron 14BNVIDIA · 15B 62Good quality (B) · 62 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking. 38.6 tok/s Runs great 55Good quality (B) · 55 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 12.2 tok/s Runs well
Qwen2.5 Coder 14B InstructAlibaba · 15B 62Good quality (B) · 62 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking. 38.6 tok/s Runs great 55Good quality (B) · 55 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 12.2 tok/s Runs well
Reka Flash 3.1 (21B)Reka AI · 21B 62Good quality (B) · 62 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking. 27.3 tok/s Runs well 46Fair quality (C) · 46 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 55/100 (quality, fit, speed and memory use) sets the ranking. 11.0 tok/s Runs well
DeepSeek Coder V2 Lite (16B)DeepSeek · 16B (2.4B active) 52Fair quality (C) · 52 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking. 82.8 tok/s Runs great 50Fair quality (C) · 50 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 41.2 tok/s Runs great
MiMo 7B RLXiaomi · 7.8B 55Good quality (B) · 55 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking. 50.5 tok/s Runs great 53Fair quality (C) · 53 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 17.8 tok/s Runs well
Ministral 3 14B ReasoningMistral AI · 14B 62Good quality (B) · 62 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking. 40.9 tok/s Runs great 62Good quality (B) · 62 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 10.0 tok/s Runs well
DeepSeek V2 Lite ChatDeepSeek · 16B (2.4B active) 40Fair quality (C) · 40 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking. 82.8 tok/s Runs great 38Basic quality (D) · 38 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 41.2 tok/s Runs great
Qwen3 Next 80B-A3B InstructAlibaba · 81B (3.0B active) 70Great quality (A) · 70 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking. 73.8 tok/s Runs welloffloads to RAM 79Great quality (A) · 79 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 85/100 (quality, fit, speed and memory use) sets the ranking. 40.2 tok/s Runs great
InternVL3 8BShanghai AI Lab · 7.9B 55Good quality (B) · 55 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking. 49.8 tok/s Runs great 53Fair quality (C) · 53 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 17.6 tok/s Runs well
Falcon 3 10B InstructTII · 10B 51Fair quality (C) · 51 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking. 38.4 tok/s Runs great 49Fair quality (C) · 49 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 13.5 tok/s Runs well
Granite 4.0 H Tiny (7B-A1B)IBM · 6.9B (1.0B active) 44Fair quality (C) · 44 of 100: how capable the model is at F16, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking. 84.4 tok/s Runs great 44Fair quality (C) · 44 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 56.0 tok/s Runs great
Fara 1.5 4BMicrosoft · 4.5B 58Good quality (B) · 58 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking. 63.9 tok/s Runs great 55Good quality (B) · 55 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 30.8 tok/s Runs great
DeepCoder 14B PreviewAgentica · 15B 60Good quality (B) · 60 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking. 38.6 tok/s Runs great 53Fair quality (C) · 53 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 12.2 tok/s Runs well
Mage-VL 4.7BMicrosoft · 4.7B 56Good quality (B) · 56 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking. 61.2 tok/s Runs great 53Fair quality (C) · 53 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 68/100 (quality, fit, speed and memory use) sets the ranking. 29.5 tok/s Runs well
Codestral 22BMistral AI · 22B 58Good quality (B) · 58 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking. 25.6 tok/s Runs well 43Fair quality (C) · 43 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 54/100 (quality, fit, speed and memory use) sets the ranking. 10.3 tok/s Runs well
Mistral Nemo 12B InstructMistral AI · 12B 50Fair quality (C) · 50 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking. 38.6 tok/s Runs great 49Fair quality (C) · 49 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. 11.4 tok/s Runs well
Llama 3.1 Nemotron Nano 8B v1NVIDIA · 8.0B 53Fair quality (C) · 53 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking. 49.3 tok/s Runs great 51Fair quality (C) · 51 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 17.4 tok/s Runs well
Yi 1.5 34B Chat01.AI · 34B 47Fair quality (C) · 47 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking. 16.6 tok/s Runs well 48Fair quality (C) · 48 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 3.4 tok/s Slow
LFM2 8B-A1BLiquid AI · 8.3B (1.5B active) 38Basic quality (D) · 38 of 100: how capable the model is at F16, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking. 64.1 tok/s Runs great 38Basic quality (D) · 38 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 40.2 tok/s Runs great
Olmo 3 7B InstructAi2 · 7.3B 50Fair quality (C) · 50 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking. 39.8 tok/s Runs great 47Fair quality (C) · 47 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 19.1 tok/s Runs well
MiniCPM4 8BOpenBMB · 8.2B 53Fair quality (C) · 53 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking. 48.3 tok/s Runs great 51Fair quality (C) · 51 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 17.1 tok/s Runs well
InternLM2.5 20B ChatShanghai AI Lab · 20B 47Fair quality (C) · 47 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking. 28.7 tok/s Runs well 35Basic quality (D) · 35 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 56/100 (quality, fit, speed and memory use) sets the ranking. 11.5 tok/s Runs well
GLM-4 9B 0414Zhipu AI · 9.4B 51Fair quality (C) · 51 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking. 42.1 tok/s Runs great 49Fair quality (C) · 49 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 14.9 tok/s Runs well
NorthCUA Nanobeipo-ai · 9.4B 51Fair quality (C) · 51 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking. 42.0 tok/s Runs great 49Fair quality (C) · 49 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 14.8 tok/s Runs well
Gemma 2 9B ITGoogle · 9.2B 49Fair quality (C) · 49 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking. 42.8 tok/s Runs great 47Fair quality (C) · 47 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. 15.1 tok/s Runs well
DeepSeek R1 Distill Qwen 14BDeepSeek · 15B 58Good quality (B) · 58 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking. 38.6 tok/s Runs great 52Fair quality (C) · 52 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 12.2 tok/s Runs well
Pixtral 12B (2409)Mistral AI · 12B 48Fair quality (C) · 48 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking. 38.0 tok/s Runs great 47Fair quality (C) · 47 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 11.3 tok/s Runs well
Command R 35B (v01)Cohere · 35B 45Fair quality (C) · 45 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking. 21.1 tok/s Runs welltight fit 52Fair quality (C) · 52 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 3.3 tok/s Slow
JanusCoder 14BShanghai AI Lab · 15B 58Good quality (B) · 58 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking. 38.6 tok/s Runs great 52Fair quality (C) · 52 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 12.2 tok/s Runs well
MiniCPM-o 2.6 8BOpenBMB · 8.7B 51Fair quality (C) · 51 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking. 45.6 tok/s Runs great 49Fair quality (C) · 49 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 16.1 tok/s Runs well
Qwen3-VL 4B InstructAlibaba · 4.4B 54Fair quality (C) · 54 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking. 65.4 tok/s Runs great 51Fair quality (C) · 51 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 68/100 (quality, fit, speed and memory use) sets the ranking. 31.5 tok/s Runs great
Qwen2.5-VL 7B InstructAlibaba · 8.3B 51Fair quality (C) · 51 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. 47.7 tok/s Runs great 49Fair quality (C) · 49 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 16.8 tok/s Runs well
Hunyuan 7B InstructTencent · 7.5B 51Fair quality (C) · 51 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. 52.7 tok/s Runs great 49Fair quality (C) · 49 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 18.6 tok/s Runs well
DeepSeek R1 0528 Qwen3 8BDeepSeek · 8.2B 61Good quality (B) · 61 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. 48.3 tok/s Runs great 58Good quality (B) · 58 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 17.1 tok/s Runs well
Qwen3 4B Instruct 2507Alibaba · 4.0B 54Fair quality (C) · 54 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. 72.2 tok/s Runs great 51Fair quality (C) · 51 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 34.7 tok/s Runs great
Qwen2.5 7B InstructAlibaba · 7.6B 51Fair quality (C) · 51 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. 51.9 tok/s Runs great 49Fair quality (C) · 49 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 18.3 tok/s Runs well
LLM4Decompile 22B V2LLM4Binary · 22B 55Good quality (B) · 55 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. 25.6 tok/s Runs well 40Fair quality (C) · 40 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 53/100 (quality, fit, speed and memory use) sets the ranking. 10.3 tok/s Runs well
InternVL2.5 8BShanghai AI Lab · 8.1B 49Fair quality (C) · 49 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. 49.0 tok/s Runs great 47Fair quality (C) · 47 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 17.3 tok/s Runs well
Foundation-Sec 1.1 8B InstructCisco Foundation AI · 8.0B 49Fair quality (C) · 49 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. 49.3 tok/s Runs great 47Fair quality (C) · 47 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 17.4 tok/s Runs well
InternLM3 8B InstructShanghai AI Lab · 8.8B 49Fair quality (C) · 49 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. 45.0 tok/s Runs great 47Fair quality (C) · 47 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. 15.9 tok/s Runs well
LFM3 1.7BLiquid AI · 1.7B 56Good quality (B) · 56 of 100: how capable the model is at F16, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. 77.3 tok/s Runs great 56Good quality (B) · 56 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. 41.1 tok/s Runs great
GLM-4 9B Chat 1MZhipu AI · 9.5B 47Fair quality (C) · 47 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. 41.7 tok/s Runs great 45Fair quality (C) · 45 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 14.7 tok/s Runs well
OLMo 2 13B InstructAi2 · 14B 41Fair quality (C) · 41 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. 41.6 tok/s Runs great 41Fair quality (C) · 41 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. 10.2 tok/s Runs well
Aya 23 35BCohere · 35B 42Fair quality (C) · 42 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. 21.1 tok/s Runs welltight fit 48Fair quality (C) · 48 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 3.3 tok/s Slow
GLM-4 9B Chat (2024)Zhipu AI · 9.4B 47Fair quality (C) · 47 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. 42.1 tok/s Runs great 45Fair quality (C) · 45 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 14.9 tok/s Runs well
GLiDE 9BFastino · 7.9B 49Fair quality (C) · 49 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. 49.8 tok/s Runs great 47Fair quality (C) · 47 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 17.6 tok/s Runs well
Qwen1.5 32B ChatAlibaba · 33B 41Fair quality (C) · 41 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. 17.5 tok/s Runs well 41Fair quality (C) · 41 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 56/100 (quality, fit, speed and memory use) sets the ranking. 4.3 tok/s Slow
Qwen2.5 7B Instruct 1MAlibaba · 7.6B 49Fair quality (C) · 49 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. 51.9 tok/s Runs great 47Fair quality (C) · 47 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 18.3 tok/s Runs well
Falcon-H1 7B InstructTII · 7.6B 49Fair quality (C) · 49 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. 52.1 tok/s Runs great 47Fair quality (C) · 47 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 18.4 tok/s Runs well
Phi-3 Medium 128K InstructMicrosoft · 14B 43Fair quality (C) · 43 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. 40.8 tok/s Runs great 43Fair quality (C) · 43 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 10.0 tok/s Runs well
Hermes 3 (Llama 3.1 8B)Nous Research · 8.0B 47Fair quality (C) · 47 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. 49.3 tok/s Runs great 45Fair quality (C) · 45 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. 17.4 tok/s Runs well
Llama 3.1 8B InstructMeta · 8.0B 47Fair quality (C) · 47 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. 49.3 tok/s Runs great 45Fair quality (C) · 45 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. 17.4 tok/s Runs well
Falcon 3 7B InstructTII · 7.5B 46Fair quality (C) · 46 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. 38.9 tok/s Runs great 43Fair quality (C) · 43 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. 18.7 tok/s Runs well
LLaVA 1.6 34BLLaVA · 35B 41Fair quality (C) · 41 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. 16.4 tok/s Runs well 41Fair quality (C) · 41 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 56/100 (quality, fit, speed and memory use) sets the ranking. 4.0 tok/s Slow
Gemma 3n E4BGoogle · 7.9B 46Fair quality (C) · 46 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. 72.6 tok/s Runs great 43Fair quality (C) · 43 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 34.9 tok/s Runs great
Granite 4.2 3BIBM · 3.7B 52Fair quality (C) · 52 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. 79.3 tok/s Runs great 49Fair quality (C) · 49 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 68/100 (quality, fit, speed and memory use) sets the ranking. 38.2 tok/s Runs great
GLM-Z1 9B (0414)Zhipu AI · 9.4B 57Good quality (B) · 57 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. 42.1 tok/s Runs great 55Good quality (B) · 55 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 14.9 tok/s Runs well
Apertus v1.5 8BSwiss AI · 8.9B 45Fair quality (C) · 45 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. 44.4 tok/s Runs great 43Fair quality (C) · 43 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 15.7 tok/s Runs well
MiniCPM-V 2.6 8BOpenBMB · 8.1B 47Fair quality (C) · 47 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. 48.8 tok/s Runs great 45Fair quality (C) · 45 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. 17.2 tok/s Runs well
Mage-VL-Cowork 4.7BMicrosoft · 4.7B 60Good quality (B) · 60 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. 61.2 tok/s Runs great 57Good quality (B) · 57 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 29.5 tok/s Runs well
Seed-Coder 8B ReasoningByteDance · 8.3B 57Good quality (B) · 57 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. 47.9 tok/s Runs great 55Good quality (B) · 55 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 16.9 tok/s Runs well
Qwen3 4BAlibaba · 4.0B 50Fair quality (C) · 50 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. 72.2 tok/s Runs great 47Fair quality (C) · 47 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 34.7 tok/s Runs great
Phi-3 Medium (14B)Microsoft · 14B 43Fair quality (C) · 43 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. 40.8 tok/s Runs great 43Fair quality (C) · 43 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 10.0 tok/s Runs well
Granite 3.2 8B InstructIBM · 8.2B 45Fair quality (C) · 45 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. 48.4 tok/s Runs great 43Fair quality (C) · 43 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 17.1 tok/s Runs well
Granite 3.3 8BIBM · 8.2B 45Fair quality (C) · 45 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. 48.4 tok/s Runs great 43Fair quality (C) · 43 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 17.1 tok/s Runs well
Ministral 8B (2410)Mistral AI · 8.0B 45Fair quality (C) · 45 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. 49.3 tok/s Runs great 43Fair quality (C) · 43 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 17.4 tok/s Runs well
Phi-3 Small 8K InstructMicrosoft · 7.4B 44Fair quality (C) · 44 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. 39.3 tok/s Runs great 41Fair quality (C) · 41 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 18.9 tok/s Runs well
DeepHermes 3 Llama 3 8B PreviewNous Research · 8.0B 45Fair quality (C) · 45 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. 49.3 tok/s Runs great 43Fair quality (C) · 43 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 17.4 tok/s Runs well
Dolphin 3.0 Llama 3.1 8BCognitive Computations · 8.0B 45Fair quality (C) · 45 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. 49.3 tok/s Runs great 43Fair quality (C) · 43 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 17.4 tok/s Runs well
Tülu 3 8BAi2 · 8.0B 45Fair quality (C) · 45 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. 49.3 tok/s Runs great 43Fair quality (C) · 43 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 17.4 tok/s Runs well
Qwen3 Coder Next (80B-A3B)Alibaba · 80B (3.0B active) 72Great quality (A) · 72 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. 75.9 tok/s Runs welloffloads to RAM 81Great quality (A) · 81 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking. 40.2 tok/s Runs great
Qwen2.5-Omni 7BAlibaba · 11B 44Fair quality (C) · 44 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. 43.9 tok/s Runs great 43Fair quality (C) · 43 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 13.0 tok/s Runs well
Olmo 3 7B ThinkAi2 · 7.3B 54Fair quality (C) · 54 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. 39.8 tok/s Runs great 51Fair quality (C) · 51 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 19.1 tok/s Runs well
EXAONE 3.5 7.8BLG AI Research · 7.8B 45Fair quality (C) · 45 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. 50.6 tok/s Runs great 43Fair quality (C) · 43 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 17.9 tok/s Runs well
CogVLM2 Llama 3 Chat 19BZhipu AI · 20B 36Basic quality (D) · 36 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. 29.2 tok/s Runs well 26Basic quality (D) · 26 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 52/100 (quality, fit, speed and memory use) sets the ranking. 11.8 tok/s Runs well
Nemotron 3 Nano 4BNVIDIA · 4.0B 50Fair quality (C) · 50 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. 73.1 tok/s Runs great 47Fair quality (C) · 47 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 35.2 tok/s Runs great
Qwen1.5 14B ChatAlibaba · 14B 32Basic quality (D) · 32 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. 40.2 tok/s Runs great 28Basic quality (D) · 28 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 53/100 (quality, fit, speed and memory use) sets the ranking. 12.7 tok/s Runs well
Seed-Coder 8B InstructByteDance · 8.3B 55Good quality (B) · 55 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. 47.9 tok/s Runs great 53Fair quality (C) · 53 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 16.9 tok/s Runs well
InternLM2 Chat 20BShanghai AI Lab · 20B 39Basic quality (D) · 39 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. 28.7 tok/s Runs well 29Basic quality (D) · 29 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 53/100 (quality, fit, speed and memory use) sets the ranking. 11.5 tok/s Runs well
Granite 3.1 8B InstructIBM · 8.2B 43Fair quality (C) · 43 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. 48.4 tok/s Runs great 41Fair quality (C) · 41 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 17.1 tok/s Runs well
EXAONE Deep 7.8BLG AI Research · 7.8B 55Good quality (B) · 55 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. 50.6 tok/s Runs great 53Fair quality (C) · 53 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 17.9 tok/s Runs well
LFM3-VL 2BLiquid AI · 1.9B 50Fair quality (C) · 50 of 100: how capable the model is at F16, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. 69.7 tok/s Runs great 49Fair quality (C) · 49 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 50.5 tok/s Runs great
Yi 34B Chat01.AI · 34B 38Basic quality (D) · 38 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. 16.6 tok/s Runs well 38Basic quality (D) · 38 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 54/100 (quality, fit, speed and memory use) sets the ranking. 4.1 tok/s Slow
InternVL2 8BShanghai AI Lab · 8.1B 43Fair quality (C) · 43 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. 49.0 tok/s Runs great 41Fair quality (C) · 41 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 17.3 tok/s Runs well
Dolphin 2.9.4 Llama 3.1 8BCognitive Computations · 8.0B 43Fair quality (C) · 43 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. 49.3 tok/s Runs great 41Fair quality (C) · 41 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 17.4 tok/s Runs well
Hermes 2 Pro Llama 3 8BNous Research · 8.0B 43Fair quality (C) · 43 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. 49.3 tok/s Runs great 41Fair quality (C) · 41 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 17.4 tok/s Runs well
OpenChat 3.6 8BOpenChat · 8.0B 43Fair quality (C) · 43 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. 49.3 tok/s Runs great 41Fair quality (C) · 41 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 17.4 tok/s Runs well
Aya Vision 8BCohere · 8.6B 43Fair quality (C) · 43 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. 45.8 tok/s Runs great 41Fair quality (C) · 41 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 16.2 tok/s Runs well
DeepSeek MoE 16B ChatDeepSeek · 16B (2.8B active) 26Basic quality (D) · 26 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. 74.3 tok/s Runs great 24Basic quality (D) · 24 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. 42.7 tok/s Runs great
Qwen2-VL 7B InstructAlibaba · 8.3B 43Fair quality (C) · 43 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. 47.7 tok/s Runs great 41Fair quality (C) · 41 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 16.8 tok/s Runs well
Phi-4 Multimodal InstructMicrosoft · 5.6B 44Fair quality (C) · 44 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. 52.1 tok/s Runs great 41Fair quality (C) · 41 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. 25.1 tok/s Runs well
Llama 3.2 11B Vision InstructMeta · 11B 40Fair quality (C) · 40 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. 44.1 tok/s Runs great 39Basic quality (D) · 39 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. 13.1 tok/s Runs well
Falcon-H1R 7BTII · 7.6B 55Good quality (B) · 55 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. 52.1 tok/s Runs great 53Fair quality (C) · 53 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 18.4 tok/s Runs well
Mistral-NeMo-Minitron 8B InstructNVIDIA · 8.4B 41Fair quality (C) · 41 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. 47.0 tok/s Runs great 39Basic quality (D) · 39 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 16.6 tok/s Runs well
Command R7BCohere · 8.0B 43Fair quality (C) · 43 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. 49.3 tok/s Runs great 41Fair quality (C) · 41 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 17.4 tok/s Runs well
Qwen2 7B InstructAlibaba · 7.6B 43Fair quality (C) · 43 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. 51.9 tok/s Runs great 41Fair quality (C) · 41 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 18.3 tok/s Runs well
Jamba Reasoning 3BAI21 Labs · 3.2B 44Fair quality (C) · 44 of 100: how capable the model is at F16, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. 41.8 tok/s Runs great 41Fair quality (C) · 41 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 43.6 tok/s Runs great
LFM2.5-VL 3BLiquid AI · 3.1B 44Fair quality (C) · 44 of 100: how capable the model is at F16, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking. 42.9 tok/s Runs great 41Fair quality (C) · 41 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 44.8 tok/s Runs great
Qwen2.5 Coder 7B InstructAlibaba · 7.6B 54Fair quality (C) · 54 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking. 51.9 tok/s Runs great 52Fair quality (C) · 52 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 18.3 tok/s Runs well
Llama 3 8B InstructMeta · 8.0B 41Fair quality (C) · 41 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking. 49.3 tok/s Runs great 39Basic quality (D) · 39 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 17.4 tok/s Runs well
Qwen3 4B Thinking 2507Alibaba · 4.0B 56Good quality (B) · 56 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking. 72.2 tok/s Runs great 53Fair quality (C) · 53 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 34.7 tok/s Runs great
Gemma 3n E2BGoogle · 5.4B 36Basic quality (D) · 36 of 100: how capable the model is at F16, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking. 66.9 tok/s Runs great 35Basic quality (D) · 35 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 48.5 tok/s Runs great
North Micro VisionCohere · 2.5B 44Fair quality (C) · 44 of 100: how capable the model is at F16, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking. 53.9 tok/s Runs great 43Fair quality (C) · 43 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 39.1 tok/s Runs great
Llama 3.1 Nemotron Nano 4B v1.1NVIDIA · 4.5B 44Fair quality (C) · 44 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking. 64.4 tok/s Runs great 41Fair quality (C) · 41 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 31.0 tok/s Runs great
OpenReasoning Nemotron 7BNVIDIA · 7.6B 53Fair quality (C) · 53 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking. 51.9 tok/s Runs great 51Fair quality (C) · 51 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 18.3 tok/s Runs well
LFM2.5 2.6BLiquid AI · 2.7B 44Fair quality (C) · 44 of 100: how capable the model is at F16, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking. 49.5 tok/s Runs great 42Fair quality (C) · 42 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 42.7 tok/s Runs great
Granite 4.1 3BIBM · 3.4B 46Fair quality (C) · 46 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking. 85.4 tok/s Runs great 43Fair quality (C) · 43 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 41.1 tok/s Runs great
SmolLM3 3BHugging Face · 3.1B 42Fair quality (C) · 42 of 100: how capable the model is at F16, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking. 43.4 tok/s Runs great 39Basic quality (D) · 39 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 45.3 tok/s Runs great
Command R7B ArabicCohere · 8.0B 41Fair quality (C) · 41 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking. 49.3 tok/s Runs great 39Basic quality (D) · 39 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 17.4 tok/s Runs well
Gemma 3 4B ITGoogle · 4.3B 45Fair quality (C) · 45 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking. 67.5 tok/s Runs great 42Fair quality (C) · 42 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 32.5 tok/s Runs great
Foundation-Sec 8B ReasoningCisco Foundation AI · 8.0B 51Fair quality (C) · 51 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking. 49.3 tok/s Runs great 49Fair quality (C) · 49 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 17.4 tok/s Runs well
Nous Hermes 2 SOLAR 10.7BNous Research · 11B 38Basic quality (D) · 38 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking. 43.9 tok/s Runs great 38Basic quality (D) · 38 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 57/100 (quality, fit, speed and memory use) sets the ranking. 13.0 tok/s Runs well
SOLAR 10.7B InstructUpstage · 11B 38Basic quality (D) · 38 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking. 43.9 tok/s Runs great 38Basic quality (D) · 38 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 57/100 (quality, fit, speed and memory use) sets the ranking. 13.0 tok/s Runs well
Phi-4 Mini (3.8B)Microsoft · 3.8B 44Fair quality (C) · 44 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking. 75.6 tok/s Runs great 41Fair quality (C) · 41 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 36.4 tok/s Runs great
Mistral 7B Instruct v0.3Mistral AI · 7.3B 38Basic quality (D) · 38 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking. 40.0 tok/s Runs great 36Basic quality (D) · 36 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. 19.3 tok/s Runs well
Ministral 3 3BMistral AI · 3.9B 44Fair quality (C) · 44 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 75.4 tok/s Runs great 41Fair quality (C) · 41 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 36.3 tok/s Runs great
Apertus 8B InstructSwiss AI · 8.1B 39Basic quality (D) · 39 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 49.1 tok/s Runs great 38Basic quality (D) · 38 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 17.3 tok/s Runs well
Aya Expanse 8BCohere · 8.0B 39Basic quality (D) · 39 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 49.3 tok/s Runs great 38Basic quality (D) · 38 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 17.4 tok/s Runs well
OLMo 2 7B InstructAi2 · 7.3B 36Basic quality (D) · 36 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 39.8 tok/s Runs great 34Basic quality (D) · 34 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 57/100 (quality, fit, speed and memory use) sets the ranking. 19.1 tok/s Runs well
Jamba2 3BAI21 Labs · 3.0B 42Fair quality (C) · 42 of 100: how capable the model is at F16, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 44.1 tok/s Runs great 40Fair quality (C) · 40 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 38.1 tok/s Runs great
MedGemma 4B ITGoogle · 4.3B 44Fair quality (C) · 44 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 67.5 tok/s Runs great 41Fair quality (C) · 41 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 32.5 tok/s Runs great
Phi-3.5 Mini InstructMicrosoft · 3.8B 40Fair quality (C) · 40 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 76.0 tok/s Runs great 38Basic quality (D) · 38 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 36.6 tok/s Runs great
InternLM2.5 7B ChatShanghai AI Lab · 7.7B 39Basic quality (D) · 39 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 51.1 tok/s Runs great 38Basic quality (D) · 38 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 18.0 tok/s Runs well
Zamba2 7B InstructZyphra · 7.4B 38Basic quality (D) · 38 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 39.4 tok/s Runs great 36Basic quality (D) · 36 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. 19.0 tok/s Runs well
Granite 3.0 8B InstructIBM · 8.2B 39Basic quality (D) · 39 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 48.4 tok/s Runs great 38Basic quality (D) · 38 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 17.1 tok/s Runs well
Hunyuan 4B InstructTencent · 4.2B 42Fair quality (C) · 42 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 68.8 tok/s Runs great 39Basic quality (D) · 39 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 33.1 tok/s Runs great
Qwen1.5 MoE A2.7B ChatAlibaba · 14B (2.7B active) 24Basic quality (D) · 24 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 76.3 tok/s Runs great 22Basic quality (D) · 22 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 57/100 (quality, fit, speed and memory use) sets the ranking. 44.1 tok/s Runs great
DeepSeek R1 Distill Llama 8BDeepSeek · 8.0B 49Fair quality (C) · 49 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 49.3 tok/s Runs great 47Fair quality (C) · 47 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. 17.4 tok/s Runs well
Falcon3 Mamba 7B InstructTII · 7.3B 38Basic quality (D) · 38 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 39.9 tok/s Runs great 36Basic quality (D) · 36 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. 19.2 tok/s Runs well
Phi-3.5 Vision InstructMicrosoft · 4.2B 38Basic quality (D) · 38 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 69.9 tok/s Runs great 36Basic quality (D) · 36 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. 33.7 tok/s Runs great
Fara 7BMicrosoft · 8.3B 49Fair quality (C) · 49 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 47.7 tok/s Runs great 47Fair quality (C) · 47 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. 16.8 tok/s Runs well
Hermes 2 Pro Mistral 7BNous Research · 7.2B 36Basic quality (D) · 36 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 40.1 tok/s Runs great 34Basic quality (D) · 34 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 57/100 (quality, fit, speed and memory use) sets the ranking. 19.3 tok/s Runs well
Starling LM 7B BetaNexusflow · 7.2B 36Basic quality (D) · 36 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 40.1 tok/s Runs great 34Basic quality (D) · 34 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 57/100 (quality, fit, speed and memory use) sets the ranking. 19.3 tok/s Runs well
Dolphin 2.9 Llama 3 8BCognitive Computations · 8.0B 37Basic quality (D) · 37 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 49.3 tok/s Runs great 36Basic quality (D) · 36 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. 17.4 tok/s Runs well
Ling Flash 2.0 (100B-A6B)Ant Group (inclusionAI) · 103B (6.1B active) 53Fair quality (C) · 53 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 68/100 (quality, fit, speed and memory use) sets the ranking. 47.9 tok/s Runs welloffloads to RAM 72Great quality (A) · 72 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 78/100 (quality, fit, speed and memory use) sets the ranking. 21.3 tok/s Runs well
OLMoE 1B-7B InstructAi2 · 6.9B (1.3B active) 26Basic quality (D) · 26 of 100: how capable the model is at F16, whatever it runs on. Match score 68/100 (quality, fit, speed and memory use) sets the ranking. 71.0 tok/s Runs great 26Basic quality (D) · 26 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. 45.3 tok/s Runs great
Hunyuan A13B (80B-A13B)Tencent · 80B (13.0B active) 62Good quality (B) · 62 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 68/100 (quality, fit, speed and memory use) sets the ranking. 20.3 tok/s Runs welloffloads to RAM 70Great quality (A) · 70 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. 10.4 tok/s Runs well
OpenChat 3.5 (0106)OpenChat · 7.2B 35Basic quality (D) · 35 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 68/100 (quality, fit, speed and memory use) sets the ranking. 40.1 tok/s Runs great 33Basic quality (D) · 33 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 57/100 (quality, fit, speed and memory use) sets the ranking. 19.3 tok/s Runs well
DeepSeek Coder 33B InstructDeepSeek · 33B 41Fair quality (C) · 41 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 68/100 (quality, fit, speed and memory use) sets the ranking. 17.1 tok/s Runs well 41Fair quality (C) · 41 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 51/100 (quality, fit, speed and memory use) sets the ranking. 4.2 tok/s Slow
Qianfan-OCR 4BBaidu · 4.7B 50Fair quality (C) · 50 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 68/100 (quality, fit, speed and memory use) sets the ranking. 61.2 tok/s Runs great 47Fair quality (C) · 47 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 29.5 tok/s Runs well
AFM 4.5BArcee AI · 4.6B 40Fair quality (C) · 40 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 68/100 (quality, fit, speed and memory use) sets the ranking. 62.8 tok/s Runs great 38Basic quality (D) · 38 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. 30.2 tok/s Runs great
EXAONE 3.0 7.8B InstructLG AI Research · 7.8B 37Basic quality (D) · 37 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 68/100 (quality, fit, speed and memory use) sets the ranking. 50.6 tok/s Runs great 36Basic quality (D) · 36 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. 17.9 tok/s Runs well
Granite 4.0 H MicroIBM · 3.2B 38Basic quality (D) · 38 of 100: how capable the model is at F16, whatever it runs on. Match score 68/100 (quality, fit, speed and memory use) sets the ranking. 41.9 tok/s Runs great 36Basic quality (D) · 36 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 43.8 tok/s Runs great
Mistral 7B Instruct v0.2Mistral AI · 7.2B 34Basic quality (D) · 34 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 68/100 (quality, fit, speed and memory use) sets the ranking. 40.1 tok/s Runs great 32Basic quality (D) · 32 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 57/100 (quality, fit, speed and memory use) sets the ranking. 19.3 tok/s Runs well
Nous Hermes 2 Mistral 7B DPONous Research · 7.2B 34Basic quality (D) · 34 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 68/100 (quality, fit, speed and memory use) sets the ranking. 40.1 tok/s Runs great 32Basic quality (D) · 32 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 57/100 (quality, fit, speed and memory use) sets the ranking. 19.3 tok/s Runs well
OpenChat 3.5 (1210)OpenChat · 7.2B 34Basic quality (D) · 34 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 68/100 (quality, fit, speed and memory use) sets the ranking. 40.1 tok/s Runs great 32Basic quality (D) · 32 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 57/100 (quality, fit, speed and memory use) sets the ranking. 19.3 tok/s Runs well
Yi 1.5 9B Chat01.AI · 8.8B 35Basic quality (D) · 35 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 68/100 (quality, fit, speed and memory use) sets the ranking. 44.8 tok/s Runs great 34Basic quality (D) · 34 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 56/100 (quality, fit, speed and memory use) sets the ranking. 15.8 tok/s Runs well
Phi-3 Mini 128K InstructMicrosoft · 3.8B 36Basic quality (D) · 36 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 68/100 (quality, fit, speed and memory use) sets the ranking. 76.0 tok/s Runs great 34Basic quality (D) · 34 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 36.6 tok/s Runs great
Qwen3.5 2BAlibaba · 2.3B 40Fair quality (C) · 40 of 100: how capable the model is at F16, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 58.9 tok/s Runs great 39Basic quality (D) · 39 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 42.7 tok/s Runs great
CodeGeeX4 All 9BZhipu AI · 9.4B 45Fair quality (C) · 45 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 42.1 tok/s Runs great 43Fair quality (C) · 43 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 55/100 (quality, fit, speed and memory use) sets the ranking. 14.9 tok/s Runs well
DeepSeek Prover V2 7BDeepSeek · 6.9B 40Fair quality (C) · 40 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 42.0 tok/s Runs great 38Basic quality (D) · 38 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 55/100 (quality, fit, speed and memory use) sets the ranking. 20.2 tok/s Runs well
Granite 4.0 Micro (3B)IBM · 3.4B 40Fair quality (C) · 40 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 85.4 tok/s Runs great 38Basic quality (D) · 38 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 41.1 tok/s Runs great
Llama 3.2 3B InstructMeta · 3.2B 35Basic quality (D) · 35 of 100: how capable the model is at F16, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 41.7 tok/s Runs great 33Basic quality (D) · 33 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 43.5 tok/s Runs great
Falcon-H1 3B InstructTII · 3.2B 36Basic quality (D) · 36 of 100: how capable the model is at F16, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 42.5 tok/s Runs great 34Basic quality (D) · 34 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. 44.3 tok/s Runs great
RecurrentGemma 9B ITGoogle · 9.6B 33Basic quality (D) · 33 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 41.1 tok/s Runs great 32Basic quality (D) · 32 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 55/100 (quality, fit, speed and memory use) sets the ranking. 14.5 tok/s Runs well
Chandra OCR 2Datalab · 5.3B 48Fair quality (C) · 48 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 54.8 tok/s Runs great 45Fair quality (C) · 45 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 26.4 tok/s Runs well
Phi-3 Vision 128K InstructMicrosoft · 4.2B 34Basic quality (D) · 34 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 69.9 tok/s Runs great 32Basic quality (D) · 32 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 33.7 tok/s Runs great
TMax 2BAi2 · 1.9B 40Fair quality (C) · 40 of 100: how capable the model is at F16, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 71.1 tok/s Runs great 39Basic quality (D) · 39 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 51.5 tok/s Runs great
OpenHermes 2.5 Mistral 7BTeknium · 7.2B 32Basic quality (D) · 32 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 40.1 tok/s Runs great 30Basic quality (D) · 30 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 56/100 (quality, fit, speed and memory use) sets the ranking. 19.3 tok/s Runs well
Starling LM 7B AlphaBerkeley · 7.2B 32Basic quality (D) · 32 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 40.1 tok/s Runs great 30Basic quality (D) · 30 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 56/100 (quality, fit, speed and memory use) sets the ranking. 19.3 tok/s Runs well
Aya 23 8BCohere · 8.0B 33Basic quality (D) · 33 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 49.3 tok/s Runs great 32Basic quality (D) · 32 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 56/100 (quality, fit, speed and memory use) sets the ranking. 17.4 tok/s Runs well
Voxtral Mini 3BMistral AI · 4.7B 36Basic quality (D) · 36 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 62.0 tok/s Runs great 34Basic quality (D) · 34 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 29.8 tok/s Runs well
MiniCPM3 4BOpenBMB · 4.1B 38Basic quality (D) · 38 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 71.3 tok/s Runs great 36Basic quality (D) · 36 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. 34.3 tok/s Runs great
S5 nanoMaritaca AI · 4.7B 37Basic quality (D) · 37 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 62.3 tok/s Runs great 35Basic quality (D) · 35 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 30.0 tok/s Runs well
DeepSeek R1 Distill Qwen 7BDeepSeek · 7.6B 45Fair quality (C) · 45 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 51.9 tok/s Runs great 43Fair quality (C) · 43 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 56/100 (quality, fit, speed and memory use) sets the ranking. 18.3 tok/s Runs well
MiniCPM-V 4.7 1BOpenBMB · 1.6B 40Fair quality (C) · 40 of 100: how capable the model is at F16, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 86.3 tok/s Runs great 40Fair quality (C) · 40 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 45.9 tok/s Runs great
Phi-3 Mini 4K InstructMicrosoft · 3.8B 36Basic quality (D) · 36 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 76.0 tok/s Runs great 34Basic quality (D) · 34 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 36.6 tok/s Runs great
Granite Code 34B InstructIBM · 34B 38Basic quality (D) · 38 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 16.9 tok/s Runs well 38Basic quality (D) · 38 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 49/100 (quality, fit, speed and memory use) sets the ranking. 4.1 tok/s Slow
LightOnOCR-3 4BLightOn · 4.5B 48Fair quality (C) · 48 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 63.9 tok/s Runs great 45Fair quality (C) · 45 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 30.8 tok/s Runs great
Gemma 1.1 7B ITGoogle · 8.5B 27Basic quality (D) · 27 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 46.3 tok/s Runs great 26Basic quality (D) · 26 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 53/100 (quality, fit, speed and memory use) sets the ranking. 16.4 tok/s Runs well
Orca 2 13BMicrosoft · 13B 26Basic quality (D) · 26 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 43.8 tok/s Runs great 26Basic quality (D) · 26 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 52/100 (quality, fit, speed and memory use) sets the ranking. 10.7 tok/s Runs well
Unlimited OCR 2 (3B-A0.6B)Baidu · 3.3B (0.6B active) 44Fair quality (C) · 44 of 100: how capable the model is at F16, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 113 tok/s Runs great 44Fair quality (C) · 44 of 100: how capable the model is at F16, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. 44.1 tok/s Runs great
Qwen2.5 3B InstructAlibaba · 3.1B 34Basic quality (D) · 34 of 100: how capable the model is at F16, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 43.3 tok/s Runs great 32Basic quality (D) · 32 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 45.2 tok/s Runs great
Qwen3-VL 2B InstructAlibaba · 2.1B 36Basic quality (D) · 36 of 100: how capable the model is at F16, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 62.8 tok/s Runs great 35Basic quality (D) · 35 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 45.5 tok/s Runs great
WeVisOne-VL 4BTencent · 4.8B 45Fair quality (C) · 45 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 60.1 tok/s Runs great 42Fair quality (C) · 42 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 28.9 tok/s Runs well
Mistral 7B Instruct v0.1Mistral AI · 7.2B 30Basic quality (D) · 30 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 40.1 tok/s Runs great 28Basic quality (D) · 28 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 55/100 (quality, fit, speed and memory use) sets the ranking. 19.3 tok/s Runs well
Hermes 3 Llama 3.2 3BNous Research · 3.2B 32Basic quality (D) · 32 of 100: how capable the model is at F16, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 41.7 tok/s Runs great 30Basic quality (D) · 30 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 43.5 tok/s Runs great
InternLM2 Math Plus 20BShanghai AI Lab · 20B 38Basic quality (D) · 38 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 28.7 tok/s Runs well 28Basic quality (D) · 28 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 48/100 (quality, fit, speed and memory use) sets the ranking. 11.5 tok/s Runs well
Baichuan 2 13B ChatBaichuan · 14B 24Basic quality (D) · 24 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 41.0 tok/s Runs great 24Basic quality (D) · 24 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 51/100 (quality, fit, speed and memory use) sets the ranking. 10.0 tok/s Runs well
Phi-4 Mini Reasoning (3.8B)Microsoft · 3.8B 46Fair quality (C) · 46 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 75.6 tok/s Runs great 43Fair quality (C) · 43 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 36.4 tok/s Runs great
Code Llama 34B InstructMeta · 34B 36Basic quality (D) · 36 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 16.9 tok/s Runs well 36Basic quality (D) · 36 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 49/100 (quality, fit, speed and memory use) sets the ranking. 4.1 tok/s Slow
Gemma 7B ITGoogle · 8.5B 25Basic quality (D) · 25 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 46.3 tok/s Runs great 21Basic quality (D) · 21 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 52/100 (quality, fit, speed and memory use) sets the ranking. 21.1 tok/s Runs well
Llama 2 13B ChatMeta · 13B 24Basic quality (D) · 24 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 43.8 tok/s Runs great 24Basic quality (D) · 24 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 51/100 (quality, fit, speed and memory use) sets the ranking. 10.7 tok/s Runs well
Qwen1.5 7B ChatAlibaba · 7.7B 25Basic quality (D) · 25 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 51.2 tok/s Runs great 21Basic quality (D) · 21 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 53/100 (quality, fit, speed and memory use) sets the ranking. 23.4 tok/s Runs well
LFM2 2.6BLiquid AI · 2.6B 34Basic quality (D) · 34 of 100: how capable the model is at F16, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 52.0 tok/s Runs great 32Basic quality (D) · 32 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 44.9 tok/s Runs great
Falcon Mamba 7B InstructTII · 7.3B 30Basic quality (D) · 30 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 39.9 tok/s Runs great 28Basic quality (D) · 28 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 55/100 (quality, fit, speed and memory use) sets the ranking. 19.2 tok/s Runs well
Neural Chat 7B v3.3Intel · 7.2B 28Basic quality (D) · 28 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 40.1 tok/s Runs great 23Basic quality (D) · 23 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 54/100 (quality, fit, speed and memory use) sets the ranking. 24.9 tok/s Runs well
DeepSeek LLM 67B ChatDeepSeek · 67B 28Basic quality (D) · 28 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 13.9 tok/s Runs welltight fit 28Basic quality (D) · 28 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 50/100 (quality, fit, speed and memory use) sets the ranking. 3.4 tok/s Slow
StableLM 2 12B ChatStability AI · 12B 27Basic quality (D) · 27 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 38.8 tok/s Runs great 26Basic quality (D) · 26 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 52/100 (quality, fit, speed and memory use) sets the ranking. 11.5 tok/s Runs well
StarCoder2 15B InstructBigCode · 16B 38Basic quality (D) · 38 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 35.7 tok/s Runs great 33Basic quality (D) · 33 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 50/100 (quality, fit, speed and memory use) sets the ranking. 11.3 tok/s Runs well
Falcon 3 3B InstructTII · 3.2B 30Basic quality (D) · 30 of 100: how capable the model is at F16, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 41.4 tok/s Runs great 28Basic quality (D) · 28 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 43.2 tok/s Runs great
Yi Coder 9B Chat01.AI · 8.8B 39Basic quality (D) · 39 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 44.8 tok/s Runs great 38Basic quality (D) · 38 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 53/100 (quality, fit, speed and memory use) sets the ranking. 15.8 tok/s Runs well
DeepSeek Coder 6.7B InstructDeepSeek · 6.7B 34Basic quality (D) · 34 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 43.1 tok/s Runs great 32Basic quality (D) · 32 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 52/100 (quality, fit, speed and memory use) sets the ranking. 20.7 tok/s Runs well
Ornith 1.5 122B-A10BDeepReinforce · 125B (10.0B active) 64Good quality (B) · 64 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 25.1 tok/s Slowheavy RAM offload 90Excellent quality (A+) · 90 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 84/100 (quality, fit, speed and memory use) sets the ranking. 11.1 tok/s Runs well
Code Llama 13B InstructMeta · 13B 28Basic quality (D) · 28 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 43.8 tok/s Runs great 28Basic quality (D) · 28 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 48/100 (quality, fit, speed and memory use) sets the ranking. 10.7 tok/s Runs well
Qwen2.5-VL 3B InstructAlibaba · 3.8B 34Basic quality (D) · 34 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 77.4 tok/s Runs great 32Basic quality (D) · 32 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 37.2 tok/s Runs great
Qwen3.5 122B-A10BAlibaba · 125B (10.0B active) 64Good quality (B) · 64 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 25.1 tok/s Slowheavy RAM offload 89Excellent quality (A+) · 89 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 84/100 (quality, fit, speed and memory use) sets the ranking. 11.1 tok/s Runs well
Unlimited OCR (3B-A0.6B)Baidu · 3.3B (0.6B active) 40Fair quality (C) · 40 of 100: how capable the model is at F16, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 113 tok/s Runs great 40Fair quality (C) · 40 of 100: how capable the model is at F16, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 44.1 tok/s Runs great
Zephyr 7B BetaHugging Face · 7.2B 26Basic quality (D) · 26 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 40.1 tok/s Runs great 21Basic quality (D) · 21 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 53/100 (quality, fit, speed and memory use) sets the ranking. 24.9 tok/s Runs well
Ring Flash 2.0 (100B-A6B)Ant Group (inclusionAI) · 103B (6.1B active) 54Fair quality (C) · 54 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 47.9 tok/s Runs welloffloads to RAM 74Great quality (A) · 74 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. 21.3 tok/s Runs well
Qwen2.5-Omni 3BAlibaba · 5.5B 30Basic quality (D) · 30 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 52.4 tok/s Runs great 25Basic quality (D) · 25 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 56/100 (quality, fit, speed and memory use) sets the ranking. 32.6 tok/s Runs great
Nemotron 3.5 Super VL 120B-A12BNVIDIA · 124B (12.0B active) 65Good quality (B) · 65 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 21.4 tok/s Slowheavy RAM offload 89Excellent quality (A+) · 89 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 83/100 (quality, fit, speed and memory use) sets the ranking. 11.2 tok/s Runs well
CodeGemma 7BGoogle · 8.5B 33Basic quality (D) · 33 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 46.3 tok/s Runs great 32Basic quality (D) · 32 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 51/100 (quality, fit, speed and memory use) sets the ranking. 16.4 tok/s Runs well
MiniCPM5 1BOpenBMB · 1.1B 36Basic quality (D) · 36 of 100: how capable the model is at F16, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 124 tok/s Runs great 36Basic quality (D) · 36 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 65.8 tok/s Runs great
Mistral Small 4 (119B-A6B)Mistral AI · 119B (6.0B active) 56Good quality (B) · 56 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 41.6 tok/s Slowheavy RAM offload 76Great quality (A) · 76 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. 21.6 tok/s Runs well
LLM4Decompile 9B V2LLM4Binary · 8.8B 37Basic quality (D) · 37 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 44.8 tok/s Runs great 36Basic quality (D) · 36 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 52/100 (quality, fit, speed and memory use) sets the ranking. 15.8 tok/s Runs well
LLaVA 1.6 Mistral 7BLLaVA · 7.6B 27Basic quality (D) · 27 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 52.3 tok/s Runs great 23Basic quality (D) · 23 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 54/100 (quality, fit, speed and memory use) sets the ranking. 23.9 tok/s Runs well
LLaVA 1.5 13BLLaVA · 13B 20Basic quality (D) · 20 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 42.7 tok/s Runs great 20Basic quality (D) · 20 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 49/100 (quality, fit, speed and memory use) sets the ranking. 10.5 tok/s Runs well
LLM4Decompile 6.7B V2LLM4Binary · 6.7B 32Basic quality (D) · 32 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 43.1 tok/s Runs great 30Basic quality (D) · 30 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 51/100 (quality, fit, speed and memory use) sets the ranking. 20.7 tok/s Runs well
Granite 3.2 2B InstructIBM · 2.5B 30Basic quality (D) · 30 of 100: how capable the model is at F16, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 52.9 tok/s Runs great 29Basic quality (D) · 29 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 45.6 tok/s Runs great
Granite 3.3 2BIBM · 2.5B 30Basic quality (D) · 30 of 100: how capable the model is at F16, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 52.9 tok/s Runs great 29Basic quality (D) · 29 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 45.6 tok/s Runs great
WizardLM 13B v1.2WizardLM · 13B 20Basic quality (D) · 20 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 43.8 tok/s Runs great 20Basic quality (D) · 20 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 49/100 (quality, fit, speed and memory use) sets the ranking. 10.7 tok/s Runs well
Llama 2 70B ChatMeta · 69B 25Basic quality (D) · 25 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 13.6 tok/s Runs welltight fit 25Basic quality (D) · 25 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 48/100 (quality, fit, speed and memory use) sets the ranking. 3.3 tok/s Slow
Ling 2.6 Flash (104B-A7.4B)Ant Group (inclusionAI) · 107B (7.4B active) 56Good quality (B) · 56 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 39.2 tok/s Slowheavy RAM offload 76Great quality (A) · 76 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. 17.7 tok/s Runs well
EXAONE 3.5 2.4BLG AI Research · 2.4B 30Basic quality (D) · 30 of 100: how capable the model is at F16, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 55.5 tok/s Runs great 29Basic quality (D) · 29 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 40.2 tok/s Runs great
Yi 1.5 6B Chat01.AI · 6.1B 28Basic quality (D) · 28 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 47.9 tok/s Runs great 23Basic quality (D) · 23 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 55/100 (quality, fit, speed and memory use) sets the ranking. 29.8 tok/s Runs well
Qwen3 1.7BAlibaba · 2.0B 30Basic quality (D) · 30 of 100: how capable the model is at F16, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 65.9 tok/s Runs great 29Basic quality (D) · 29 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 47.7 tok/s Runs great
Granite Code 20B InstructIBM · 20B 34Basic quality (D) · 34 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 28.4 tok/s Runs well 25Basic quality (D) · 25 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 46/100 (quality, fit, speed and memory use) sets the ranking. 11.4 tok/s Runs well
Codestral Mamba 7BMistral AI · 7.3B 36Basic quality (D) · 36 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 39.8 tok/s Runs great 34Basic quality (D) · 34 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 52/100 (quality, fit, speed and memory use) sets the ranking. 19.2 tok/s Runs well
SQLCoder 7B v2Defog · 6.7B 30Basic quality (D) · 30 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 43.1 tok/s Runs great 28Basic quality (D) · 28 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 50/100 (quality, fit, speed and memory use) sets the ranking. 20.7 tok/s Runs well
Granite 3.1 2B InstructIBM · 2.5B 28Basic quality (D) · 28 of 100: how capable the model is at F16, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 52.9 tok/s Runs great 27Basic quality (D) · 27 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 45.6 tok/s Runs great
Gemma 2 2BGoogle · 2.6B 28Basic quality (D) · 28 of 100: how capable the model is at F16, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 51.2 tok/s Runs great 27Basic quality (D) · 27 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 44.2 tok/s Runs great
Vicuna 13B v1.5LMSYS · 13B 19Basic quality (D) · 19 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 43.8 tok/s Runs great 16Basic quality (D) · 16 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 48/100 (quality, fit, speed and memory use) sets the ranking. 13.9 tok/s Runs well
Mathstral 7BMistral AI · 7.3B 34Basic quality (D) · 34 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 40.0 tok/s Runs great 32Basic quality (D) · 32 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 52/100 (quality, fit, speed and memory use) sets the ranking. 19.3 tok/s Runs well
Orca 2 7BMicrosoft · 6.7B 22Basic quality (D) · 22 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 43.1 tok/s Runs great 15Basic quality (D) · 15 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 52/100 (quality, fit, speed and memory use) sets the ranking. 34.0 tok/s Runs great
Granite Vision 3.2 2BIBM · 3.0B 26Basic quality (D) · 26 of 100: how capable the model is at F16, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 44.9 tok/s Runs great 24Basic quality (D) · 24 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. 46.9 tok/s Runs great
WeVisOne VL 2BTencent · 2.4B 27Basic quality (D) · 27 of 100: how capable the model is at F16, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. 54.8 tok/s Runs great 26Basic quality (D) · 26 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. 39.7 tok/s Runs great
LFM2.5 1.2B InstructLiquid AI · 1.2B 32Basic quality (D) · 32 of 100: how capable the model is at F16, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. 114 tok/s Runs great 32Basic quality (D) · 32 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 60.8 tok/s Runs great
Nemotron 3 Super 120B-A12BNVIDIA · 124B (12.0B active) 61Good quality (B) · 61 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. 21.5 tok/s Slowheavy RAM offload 83Great quality (A) · 83 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking. 11.2 tok/s Runs well
Qwen2.5 Coder 3B InstructAlibaba · 3.1B 36Basic quality (D) · 36 of 100: how capable the model is at F16, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. 43.3 tok/s Runs great 34Basic quality (D) · 34 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 57/100 (quality, fit, speed and memory use) sets the ranking. 45.2 tok/s Runs great
Qwen2.5 72B InstructAlibaba · 73B 61Good quality (B) · 61 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. 4.2 tok/s Slowoffloads to RAM 51Fair quality (C) · 51 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 3.2 tok/s Slow
MiniCPM-OCR 1.0 1.3BOpenBMB · 1.3B 42Fair quality (C) · 42 of 100: how capable the model is at F16, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. 103 tok/s Runs great 42Fair quality (C) · 42 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 54.7 tok/s Runs great
Qwen2.5-VL 72B InstructAlibaba · 73B 61Good quality (B) · 61 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. 4.1 tok/s Slowoffloads to RAM 51Fair quality (C) · 51 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 3.1 tok/s Slow
InternVL3 78BShanghai AI Lab · 78B 62Good quality (B) · 62 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. 3.4 tok/s Slowoffloads to RAM 73Great quality (A) · 73 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking. 1.2 tok/s Very slow
Granite 3.0 2B InstructIBM · 2.6B 26Basic quality (D) · 26 of 100: how capable the model is at F16, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. 50.9 tok/s Runs great 25Basic quality (D) · 25 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. 43.9 tok/s Runs great
TeleOCRXingChen-AGI · 1.4B 40Fair quality (C) · 40 of 100: how capable the model is at F16, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. 94.2 tok/s Runs great 40Fair quality (C) · 40 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 50.1 tok/s Runs great
Qwen2.5-Math 7B InstructAlibaba · 7.6B 35Basic quality (D) · 35 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. 51.9 tok/s Runs great 34Basic quality (D) · 34 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 52/100 (quality, fit, speed and memory use) sets the ranking. 18.3 tok/s Runs well
EXAONE 4.0 1.2BLG AI Research · 1.3B 30Basic quality (D) · 30 of 100: how capable the model is at F16, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. 105 tok/s Runs great 30Basic quality (D) · 30 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 55.5 tok/s Runs great
Llama 2 7B ChatMeta · 6.7B 20Basic quality (D) · 20 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. 43.1 tok/s Runs great 14Basic quality (D) · 14 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 52/100 (quality, fit, speed and memory use) sets the ranking. 34.0 tok/s Runs great
Leanstral 119B-A6BMistral AI · 119B (6.5B active) 51Fair quality (C) · 51 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. 38.7 tok/s Slowheavy RAM offload 70Great quality (A) · 70 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking. 20.0 tok/s Runs well
PaddleOCR-VL 1.6 0.9BBaidu · 1.0B 42Fair quality (C) · 42 of 100: how capable the model is at F16, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 139 tok/s Runs great 42Fair quality (C) · 42 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 74.1 tok/s Runs great
StarCoder2 15B (base)BigCode · 16B 30Basic quality (D) · 30 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 35.7 tok/s Runs great 26Basic quality (D) · 26 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 47/100 (quality, fit, speed and memory use) sets the ranking. 11.3 tok/s Runs well
SmolVLM2 2.2B InstructHugging Face · 2.3B 24Basic quality (D) · 24 of 100: how capable the model is at F16, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 59.4 tok/s Runs great 23Basic quality (D) · 23 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 57/100 (quality, fit, speed and memory use) sets the ranking. 43.1 tok/s Runs great
GLM-4.6VZhipu AI · 108B (12.0B active) 58Good quality (B) · 58 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 23.7 tok/s Slowheavy RAM offload 78Great quality (A) · 78 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 78/100 (quality, fit, speed and memory use) sets the ranking. 11.2 tok/s Runs well
Code Llama 70B InstructMeta · 69B 30Basic quality (D) · 30 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 13.6 tok/s Runs welltight fit 30Basic quality (D) · 30 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 46/100 (quality, fit, speed and memory use) sets the ranking. 3.3 tok/s Slow
Baichuan 2 7B ChatBaichuan · 7.5B 19Basic quality (D) · 19 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 52.7 tok/s Runs great 14Basic quality (D) · 14 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 51/100 (quality, fit, speed and memory use) sets the ranking. 30.5 tok/s Runs great
Granite Code 8B Instruct (128K)IBM · 8.1B 31Basic quality (D) · 31 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 49.1 tok/s Runs great 30Basic quality (D) · 30 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 50/100 (quality, fit, speed and memory use) sets the ranking. 17.3 tok/s Runs well
DeepSeek LLM 7B ChatDeepSeek · 6.9B 18Basic quality (D) · 18 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 42.0 tok/s Runs great 12Basic quality (D) · 12 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 51/100 (quality, fit, speed and memory use) sets the ranking. 33.2 tok/s Runs great
Granite 4.0 H 1BIBM · 1.5B 28Basic quality (D) · 28 of 100: how capable the model is at F16, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 91.6 tok/s Runs great 28Basic quality (D) · 28 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 48.7 tok/s Runs great
InternLM2 Math Plus 7BShanghai AI Lab · 7.7B 31Basic quality (D) · 31 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 51.1 tok/s Runs great 30Basic quality (D) · 30 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 50/100 (quality, fit, speed and memory use) sets the ranking. 18.0 tok/s Runs well
Hermes 4 70BNous Research · 71B 51Fair quality (C) · 51 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 10.8 tok/s Runs welloffloads to RAM 51Fair quality (C) · 51 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 3.3 tok/s Slow
Llama 3.3 70B InstructMeta · 71B 51Fair quality (C) · 51 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 10.7 tok/s Runs welloffloads to RAM 51Fair quality (C) · 51 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 3.2 tok/s Slow
Solar Open 100BUpstage · 103B (12.0B active) 56Good quality (B) · 56 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 25.1 tok/s Slowheavy RAM offload 76Great quality (A) · 76 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking. 11.2 tok/s Runs well
Groot 65B A6B 261023GrootTeam · 118B (6.0B active) 49Fair quality (C) · 49 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 41.9 tok/s Slowheavy RAM offload 66Good quality (B) · 66 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking. 21.6 tok/s Runs well
SmolVLM 2.2B InstructHugging Face · 2.3B 22Basic quality (D) · 22 of 100: how capable the model is at F16, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 59.4 tok/s Runs great 21Basic quality (D) · 21 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 56/100 (quality, fit, speed and memory use) sets the ranking. 43.1 tok/s Runs great
Qwen2-VL 2B InstructAlibaba · 2.2B 24Basic quality (D) · 24 of 100: how capable the model is at F16, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 60.5 tok/s Runs great 23Basic quality (D) · 23 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 57/100 (quality, fit, speed and memory use) sets the ranking. 43.8 tok/s Runs great
GLM-4.5VZhipu AI · 108B (12.0B active) 56Good quality (B) · 56 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 23.7 tok/s Slowheavy RAM offload 76Great quality (A) · 76 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking. 11.2 tok/s Runs well
Granite 3.1 3B-A800M InstructIBM · 3.3B (0.8B active) 20Basic quality (D) · 20 of 100: how capable the model is at F16, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 96.6 tok/s Runs great 20Basic quality (D) · 20 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 56/100 (quality, fit, speed and memory use) sets the ranking. 66.4 tok/s Runs great
LLaVA 1.5 7BLLaVA · 7.1B 16Basic quality (D) · 16 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 41.1 tok/s Runs great 11Basic quality (D) · 11 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 50/100 (quality, fit, speed and memory use) sets the ranking. 32.5 tok/s Runs great
Falcon-H1 1.5B InstructTII · 1.6B 26Basic quality (D) · 26 of 100: how capable the model is at F16, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 86.3 tok/s Runs great 26Basic quality (D) · 26 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. 45.9 tok/s Runs great
Sarvam 105BSarvam AI · 106B (10.0B active) 52Fair quality (C) · 52 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 30.1 tok/s Slowheavy RAM offload 71Great quality (A) · 71 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking. 13.3 tok/s Runs well
Code Llama 7B InstructMeta · 6.7B 24Basic quality (D) · 24 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 43.1 tok/s Runs great 16Basic quality (D) · 16 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 48/100 (quality, fit, speed and memory use) sets the ranking. 34.0 tok/s Runs great
Zamba2 2.7B InstructZyphra · 2.7B 22Basic quality (D) · 22 of 100: how capable the model is at F16, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 50.3 tok/s Runs great 21Basic quality (D) · 21 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 56/100 (quality, fit, speed and memory use) sets the ranking. 43.4 tok/s Runs great
QVQ 72B PreviewAlibaba · 73B 57Good quality (B) · 57 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 4.1 tok/s Slowoffloads to RAM 47Fair quality (C) · 47 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 3.1 tok/s Slow
DeepSeek R1 Distill Llama 70BDeepSeek · 71B 50Fair quality (C) · 50 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 10.8 tok/s Runs welloffloads to RAM 50Fair quality (C) · 50 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 3.3 tok/s Slow
Tülu 3 70BAi2 · 71B 50Fair quality (C) · 50 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 10.8 tok/s Runs welloffloads to RAM 50Fair quality (C) · 50 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 3.3 tok/s Slow
Yi 6B Chat01.AI · 6.1B 20Basic quality (D) · 20 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 47.9 tok/s Runs great 14Basic quality (D) · 14 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 52/100 (quality, fit, speed and memory use) sets the ranking. 37.9 tok/s Runs great
Falcon 40B InstructTII · 42B 11Basic quality (D) · 11 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 22.4 tok/s Runs well 15Basic quality (D) · 15 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 44/100 (quality, fit, speed and memory use) sets the ranking. 3.3 tok/s Slow
Kimi Dev 72BMoonshot AI · 73B 67Good quality (B) · 67 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 4.2 tok/s Slowoffloads to RAM 56Good quality (B) · 56 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 57/100 (quality, fit, speed and memory use) sets the ranking. 3.2 tok/s Slow
Nemotron Mini 4B InstructNVIDIA · 4.2B 22Basic quality (D) · 22 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 69.3 tok/s Runs great 18Basic quality (D) · 18 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 55/100 (quality, fit, speed and memory use) sets the ranking. 43.1 tok/s Runs great
LFM2 1.2BLiquid AI · 1.2B 26Basic quality (D) · 26 of 100: how capable the model is at F16, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 114 tok/s Runs great 26Basic quality (D) · 26 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. 60.8 tok/s Runs great
GLM-4.5 Air (106B-A12B)Zhipu AI · 110B (12.0B active) 54Fair quality (C) · 54 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 23.0 tok/s Slowheavy RAM offload 74Great quality (A) · 74 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking. 11.2 tok/s Runs well
Apertus v1.5 70BSwiss AI · 72B 55Good quality (B) · 55 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 4.3 tok/s Slowoffloads to RAM 46Fair quality (C) · 46 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. 3.2 tok/s Slow
Llama 3.1 70B InstructMeta · 71B 49Fair quality (C) · 49 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 10.7 tok/s Runs welloffloads to RAM 49Fair quality (C) · 49 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 3.2 tok/s Slow
Qwen2.5 1.5B InstructAlibaba · 1.5B 24Basic quality (D) · 24 of 100: how capable the model is at F16, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 86.9 tok/s Runs great 24Basic quality (D) · 24 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 57/100 (quality, fit, speed and memory use) sets the ranking. 46.2 tok/s Runs great
Granite 3.0 3B-A800M InstructIBM · 3.4B (0.8B active) 18Basic quality (D) · 18 of 100: how capable the model is at F16, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 96.6 tok/s Runs great 18Basic quality (D) · 18 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 55/100 (quality, fit, speed and memory use) sets the ranking. 66.4 tok/s Runs great
LFM2-VL 1.6BLiquid AI · 1.6B 24Basic quality (D) · 24 of 100: how capable the model is at F16, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 84.7 tok/s Runs great 24Basic quality (D) · 24 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 57/100 (quality, fit, speed and memory use) sets the ranking. 45.0 tok/s Runs great
Qwen3.5 0.8BAlibaba · 0.9B 26Basic quality (D) · 26 of 100: how capable the model is at F16, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 154 tok/s Runs great 26Basic quality (D) · 26 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. 81.7 tok/s Runs great
WizardMath 7B v1.1WizardLM · 7.2B 26Basic quality (D) · 26 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 40.1 tok/s Runs great 21Basic quality (D) · 21 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 48/100 (quality, fit, speed and memory use) sets the ranking. 24.9 tok/s Runs well
Vicuna 7B v1.5LMSYS · 6.7B 14Basic quality (D) · 14 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 43.1 tok/s Runs great 9Basic quality (D) · 9 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 50/100 (quality, fit, speed and memory use) sets the ranking. 34.0 tok/s Runs great
Hunyuan 1.8B InstructTencent · 1.8B 22Basic quality (D) · 22 of 100: how capable the model is at F16, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 74.7 tok/s Runs great 22Basic quality (D) · 22 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 56/100 (quality, fit, speed and memory use) sets the ranking. 39.7 tok/s Runs great
Laguna S 2.1 (118B-A8B)Poolside · 118B (8.0B active) 60Good quality (B) · 60 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 31.2 tok/s Slowheavy RAM offload 81Great quality (A) · 81 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking. 16.5 tok/s Runs well
SmolLM2 1.7B InstructHugging Face · 1.7B 20Basic quality (D) · 20 of 100: how capable the model is at F16, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 78.2 tok/s Runs great 20Basic quality (D) · 20 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 56/100 (quality, fit, speed and memory use) sets the ranking. 41.6 tok/s Runs great
Hermes 3 (Llama 3.1 70B)Nous Research · 71B 47Fair quality (C) · 47 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 10.8 tok/s Runs welloffloads to RAM 47Fair quality (C) · 47 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 3.3 tok/s Slow
Qwen2 72B InstructAlibaba · 73B 53Fair quality (C) · 53 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 4.2 tok/s Slowoffloads to RAM 44Fair quality (C) · 44 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 57/100 (quality, fit, speed and memory use) sets the ranking. 3.2 tok/s Slow
StarCoder 15BBigCode · 16B 24Basic quality (D) · 24 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 36.0 tok/s Runs great 21Basic quality (D) · 21 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 45/100 (quality, fit, speed and memory use) sets the ranking. 11.4 tok/s Runs well
Qwen2-VL 72B InstructAlibaba · 73B 53Fair quality (C) · 53 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. 4.1 tok/s Slowoffloads to RAM 44Fair quality (C) · 44 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 57/100 (quality, fit, speed and memory use) sets the ranking. 3.1 tok/s Slow
OLMo 7B InstructAi2 · 6.9B 14Basic quality (D) · 14 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. 42.1 tok/s Runs great 9Basic quality (D) · 9 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 50/100 (quality, fit, speed and memory use) sets the ranking. 33.3 tok/s Runs great
StableLM Zephyr 3BStability AI · 2.8B 16Basic quality (D) · 16 of 100: how capable the model is at F16, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. 47.8 tok/s Runs great 15Basic quality (D) · 15 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 54/100 (quality, fit, speed and memory use) sets the ranking. 41.2 tok/s Runs great
Llama 3.1 Nemotron 70B InstructNVIDIA · 71B 46Fair quality (C) · 46 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. 10.8 tok/s Runs welloffloads to RAM 46Fair quality (C) · 46 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. 3.3 tok/s Slow
Llama 3.2 1B InstructMeta · 1.2B 22Basic quality (D) · 22 of 100: how capable the model is at F16, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. 108 tok/s Runs great 22Basic quality (D) · 22 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 56/100 (quality, fit, speed and memory use) sets the ranking. 57.3 tok/s Runs great
Phi-2 2.7BMicrosoft · 2.8B 16Basic quality (D) · 16 of 100: how capable the model is at F16, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. 48.1 tok/s Runs great 15Basic quality (D) · 15 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 53/100 (quality, fit, speed and memory use) sets the ranking. 41.5 tok/s Runs great
StarCoder2 7BBigCode · 7.2B 24Basic quality (D) · 24 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. 40.5 tok/s Runs great 16Basic quality (D) · 16 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 48/100 (quality, fit, speed and memory use) sets the ranking. 32.0 tok/s Runs great
InternLM2.5 1.8B ChatShanghai AI Lab · 1.9B 18Basic quality (D) · 18 of 100: how capable the model is at F16, whatever it runs on. Match score 57/100 (quality, fit, speed and memory use) sets the ranking. 70.8 tok/s Runs great 17Basic quality (D) · 17 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 55/100 (quality, fit, speed and memory use) sets the ranking. 51.3 tok/s Runs great
Gemma 3 1BGoogle · 1.0B 20Basic quality (D) · 20 of 100: how capable the model is at F16, whatever it runs on. Match score 57/100 (quality, fit, speed and memory use) sets the ranking. 134 tok/s Runs great 20Basic quality (D) · 20 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 55/100 (quality, fit, speed and memory use) sets the ranking. 71.1 tok/s Runs great
InternLM 7BShanghai AI Lab · 7.3B 11Basic quality (D) · 11 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 57/100 (quality, fit, speed and memory use) sets the ranking. 54.0 tok/s Runs great 8Basic quality (D) · 8 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 49/100 (quality, fit, speed and memory use) sets the ranking. 31.3 tok/s Runs great
GLiDE 2BFastino · 1.9B 17Basic quality (D) · 17 of 100: how capable the model is at F16, whatever it runs on. Match score 57/100 (quality, fit, speed and memory use) sets the ranking. 71.1 tok/s Runs great 16Basic quality (D) · 16 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 54/100 (quality, fit, speed and memory use) sets the ranking. 51.5 tok/s Runs great
Apertus 70B InstructSwiss AI · 71B 42Fair quality (C) · 42 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 56/100 (quality, fit, speed and memory use) sets the ranking. 10.7 tok/s Runs welloffloads to RAM 42Fair quality (C) · 42 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 56/100 (quality, fit, speed and memory use) sets the ranking. 3.2 tok/s Slow
Falcon 3 1B InstructTII · 1.7B 16Basic quality (D) · 16 of 100: how capable the model is at F16, whatever it runs on. Match score 56/100 (quality, fit, speed and memory use) sets the ranking. 80.1 tok/s Runs great 16Basic quality (D) · 16 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 54/100 (quality, fit, speed and memory use) sets the ranking. 42.6 tok/s Runs great
Stable Code Instruct 3BStability AI · 2.8B 20Basic quality (D) · 20 of 100: how capable the model is at F16, whatever it runs on. Match score 56/100 (quality, fit, speed and memory use) sets the ranking. 47.8 tok/s Runs great 19Basic quality (D) · 19 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 50/100 (quality, fit, speed and memory use) sets the ranking. 41.2 tok/s Runs great
Llama 4 Scout 17B-16EMeta · 109B (17.0B active) 50Fair quality (C) · 50 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 56/100 (quality, fit, speed and memory use) sets the ranking. 16.9 tok/s Slowheavy RAM offload 60Good quality (B) · 60 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 10.3 tok/s Runs well
Spark X2.6 1.7BSparkLLM · 1.7B 16Basic quality (D) · 16 of 100: how capable the model is at F16, whatever it runs on. Match score 56/100 (quality, fit, speed and memory use) sets the ranking. 78.2 tok/s Runs great 16Basic quality (D) · 16 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 54/100 (quality, fit, speed and memory use) sets the ranking. 41.6 tok/s Runs great
Qwen2 1.5B InstructAlibaba · 1.5B 16Basic quality (D) · 16 of 100: how capable the model is at F16, whatever it runs on. Match score 56/100 (quality, fit, speed and memory use) sets the ranking. 86.9 tok/s Runs great 16Basic quality (D) · 16 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 54/100 (quality, fit, speed and memory use) sets the ranking. 46.2 tok/s Runs great
Llama 3 70B InstructMeta · 71B 40Fair quality (C) · 40 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 56/100 (quality, fit, speed and memory use) sets the ranking. 10.8 tok/s Runs welloffloads to RAM 40Fair quality (C) · 40 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 55/100 (quality, fit, speed and memory use) sets the ranking. 3.3 tok/s Slow
DeepScaleR 1.5B PreviewAgentica · 1.8B 26Basic quality (D) · 26 of 100: how capable the model is at F16, whatever it runs on. Match score 56/100 (quality, fit, speed and memory use) sets the ranking. 75.1 tok/s Runs great 26Basic quality (D) · 26 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 53/100 (quality, fit, speed and memory use) sets the ranking. 39.9 tok/s Runs great
LFM2 700MLiquid AI · 0.7B 18Basic quality (D) · 18 of 100: how capable the model is at F16, whatever it runs on. Match score 55/100 (quality, fit, speed and memory use) sets the ranking. 181 tok/s Runs great 18Basic quality (D) · 18 of 100: how capable the model is at F16, whatever it runs on. Match score 55/100 (quality, fit, speed and memory use) sets the ranking. 44.3 tok/s Runs great
Gemma 2B ITGoogle · 2.5B 12Basic quality (D) · 12 of 100: how capable the model is at F16, whatever it runs on. Match score 55/100 (quality, fit, speed and memory use) sets the ranking. 53.3 tok/s Runs great 11Basic quality (D) · 11 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 52/100 (quality, fit, speed and memory use) sets the ranking. 46.0 tok/s Runs great
Qwen2.5 Coder 1.5B InstructAlibaba · 1.5B 26Basic quality (D) · 26 of 100: how capable the model is at F16, whatever it runs on. Match score 55/100 (quality, fit, speed and memory use) sets the ranking. 86.9 tok/s Runs great 26Basic quality (D) · 26 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 53/100 (quality, fit, speed and memory use) sets the ranking. 46.2 tok/s Runs great
Llama 3.2 90B Vision InstructMeta · 89B 46Fair quality (C) · 46 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 55/100 (quality, fit, speed and memory use) sets the ranking. 4.2 tok/s Slowoffloads to RAM 65Good quality (B) · 65 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 1.1 tok/s Very slow
SmolLM 1.7B InstructHugging Face · 1.7B 14Basic quality (D) · 14 of 100: how capable the model is at F16, whatever it runs on. Match score 55/100 (quality, fit, speed and memory use) sets the ranking. 78.2 tok/s Runs great 14Basic quality (D) · 14 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 53/100 (quality, fit, speed and memory use) sets the ranking. 41.6 tok/s Runs great
OLMo 2 1B InstructAi2 · 1.5B 14Basic quality (D) · 14 of 100: how capable the model is at F16, whatever it runs on. Match score 55/100 (quality, fit, speed and memory use) sets the ranking. 90.4 tok/s Runs great 14Basic quality (D) · 14 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 53/100 (quality, fit, speed and memory use) sets the ranking. 48.0 tok/s Runs great
Granite Code 3B Instruct (128K)IBM · 3.5B 20Basic quality (D) · 20 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 55/100 (quality, fit, speed and memory use) sets the ranking. 83.4 tok/s Runs great 19Basic quality (D) · 19 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 50/100 (quality, fit, speed and memory use) sets the ranking. 40.1 tok/s Runs great
Falcon 7B InstructTII · 7.2B 6Basic quality (D) · 6 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 54/100 (quality, fit, speed and memory use) sets the ranking. 40.2 tok/s Runs great 4Basic quality (D) · 4 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 47/100 (quality, fit, speed and memory use) sets the ranking. 31.8 tok/s Runs great
Qwen3 0.6BAlibaba · 0.8B 14Basic quality (D) · 14 of 100: how capable the model is at F16, whatever it runs on. Match score 54/100 (quality, fit, speed and memory use) sets the ranking. 178 tok/s Runs great 14Basic quality (D) · 14 of 100: how capable the model is at F16, whatever it runs on. Match score 53/100 (quality, fit, speed and memory use) sets the ranking. 43.7 tok/s Runs great
StarCoder2 3BBigCode · 3.0B 18Basic quality (D) · 18 of 100: how capable the model is at F16, whatever it runs on. Match score 54/100 (quality, fit, speed and memory use) sets the ranking. 44.1 tok/s Runs great 17Basic quality (D) · 17 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 49/100 (quality, fit, speed and memory use) sets the ranking. 46.1 tok/s Runs great
DeepSeek R1 Distill Qwen 1.5BDeepSeek · 1.8B 22Basic quality (D) · 22 of 100: how capable the model is at F16, whatever it runs on. Match score 54/100 (quality, fit, speed and memory use) sets the ranking. 75.1 tok/s Runs great 22Basic quality (D) · 22 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 51/100 (quality, fit, speed and memory use) sets the ranking. 39.9 tok/s Runs great
StableLM 2 1.6B ChatStability AI · 1.6B 10Basic quality (D) · 10 of 100: how capable the model is at F16, whatever it runs on. Match score 54/100 (quality, fit, speed and memory use) sets the ranking. 81.6 tok/s Runs great 10Basic quality (D) · 10 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 51/100 (quality, fit, speed and memory use) sets the ranking. 43.4 tok/s Runs great
StableLM 2 Zephyr 1.6BStability AI · 1.6B 10Basic quality (D) · 10 of 100: how capable the model is at F16, whatever it runs on. Match score 54/100 (quality, fit, speed and memory use) sets the ranking. 81.6 tok/s Runs great 10Basic quality (D) · 10 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 51/100 (quality, fit, speed and memory use) sets the ranking. 43.4 tok/s Runs great
d1 2BLiquid AI · 1.9B 21Basic quality (D) · 21 of 100: how capable the model is at F16, whatever it runs on. Match score 53/100 (quality, fit, speed and memory use) sets the ranking. 69.7 tok/s Runs great 20Basic quality (D) · 20 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 51/100 (quality, fit, speed and memory use) sets the ranking. 50.5 tok/s Runs great
Falcon-H1 0.5B InstructTII · 0.5B 14Basic quality (D) · 14 of 100: how capable the model is at F16, whatever it runs on. Match score 53/100 (quality, fit, speed and memory use) sets the ranking. 257 tok/s Runs great 14Basic quality (D) · 14 of 100: how capable the model is at F16, whatever it runs on. Match score 53/100 (quality, fit, speed and memory use) sets the ranking. 63.0 tok/s Runs great
LFM2.5 350MLiquid AI · 0.4B 14Basic quality (D) · 14 of 100: how capable the model is at F16, whatever it runs on. Match score 53/100 (quality, fit, speed and memory use) sets the ranking. 382 tok/s Runs great 14Basic quality (D) · 14 of 100: how capable the model is at F16, whatever it runs on. Match score 53/100 (quality, fit, speed and memory use) sets the ranking. 93.6 tok/s Runs great
Command A (111B)Cohere · 111B 56Good quality (B) · 56 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 53/100 (quality, fit, speed and memory use) sets the ranking. 2.8 tok/s Very slowheavy RAM offload 80Great quality (A) · 80 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking. 0.6 tok/s Very slow
Qwen2.5-Math 1.5B InstructAlibaba · 1.5B 20Basic quality (D) · 20 of 100: how capable the model is at F16, whatever it runs on. Match score 52/100 (quality, fit, speed and memory use) sets the ranking. 86.9 tok/s Runs great 20Basic quality (D) · 20 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 50/100 (quality, fit, speed and memory use) sets the ranking. 46.2 tok/s Runs great
LFM2-VL 450MLiquid AI · 0.5B 12Basic quality (D) · 12 of 100: how capable the model is at F16, whatever it runs on. Match score 52/100 (quality, fit, speed and memory use) sets the ranking. 297 tok/s Runs great 12Basic quality (D) · 12 of 100: how capable the model is at F16, whatever it runs on. Match score 52/100 (quality, fit, speed and memory use) sets the ranking. 72.8 tok/s Runs great
GLiDE 0.8BFastino · 0.8B 11Basic quality (D) · 11 of 100: how capable the model is at F16, whatever it runs on. Match score 52/100 (quality, fit, speed and memory use) sets the ranking. 178 tok/s Runs great 11Basic quality (D) · 11 of 100: how capable the model is at F16, whatever it runs on. Match score 51/100 (quality, fit, speed and memory use) sets the ranking. 43.7 tok/s Runs great
LFM2 350MLiquid AI · 0.4B 12Basic quality (D) · 12 of 100: how capable the model is at F16, whatever it runs on. Match score 52/100 (quality, fit, speed and memory use) sets the ranking. 382 tok/s Runs great 12Basic quality (D) · 12 of 100: how capable the model is at F16, whatever it runs on. Match score 52/100 (quality, fit, speed and memory use) sets the ranking. 93.6 tok/s Runs great
Granite 4.0 H 350MIBM · 0.3B 12Basic quality (D) · 12 of 100: how capable the model is at F16, whatever it runs on. Match score 52/100 (quality, fit, speed and memory use) sets the ranking. 393 tok/s Runs great 12Basic quality (D) · 12 of 100: how capable the model is at F16, whatever it runs on. Match score 52/100 (quality, fit, speed and memory use) sets the ranking. 96.4 tok/s Runs great
Yi Coder 1.5B Chat01.AI · 1.5B 16Basic quality (D) · 16 of 100: how capable the model is at F16, whatever it runs on. Match score 52/100 (quality, fit, speed and memory use) sets the ranking. 90.4 tok/s Runs great 16Basic quality (D) · 16 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 49/100 (quality, fit, speed and memory use) sets the ranking. 48.0 tok/s Runs great
SmolVLM 500M InstructHugging Face · 0.5B 10Basic quality (D) · 10 of 100: how capable the model is at F16, whatever it runs on. Match score 52/100 (quality, fit, speed and memory use) sets the ranking. 262 tok/s Runs great 10Basic quality (D) · 10 of 100: how capable the model is at F16, whatever it runs on. Match score 51/100 (quality, fit, speed and memory use) sets the ranking. 64.3 tok/s Runs great
Qwen2.5 0.5B InstructAlibaba · 0.5B 10Basic quality (D) · 10 of 100: how capable the model is at F16, whatever it runs on. Match score 51/100 (quality, fit, speed and memory use) sets the ranking. 273 tok/s Runs great 10Basic quality (D) · 10 of 100: how capable the model is at F16, whatever it runs on. Match score 51/100 (quality, fit, speed and memory use) sets the ranking. 66.9 tok/s Runs great
CodeGemma 2BGoogle · 2.5B 14Basic quality (D) · 14 of 100: how capable the model is at F16, whatever it runs on. Match score 51/100 (quality, fit, speed and memory use) sets the ranking. 53.3 tok/s Runs great 13Basic quality (D) · 13 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 48/100 (quality, fit, speed and memory use) sets the ranking. 46.0 tok/s Runs great
Hunyuan 0.5B InstructTencent · 0.5B 8Basic quality (D) · 8 of 100: how capable the model is at F16, whatever it runs on. Match score 51/100 (quality, fit, speed and memory use) sets the ranking. 248 tok/s Runs great 8Basic quality (D) · 8 of 100: how capable the model is at F16, whatever it runs on. Match score 50/100 (quality, fit, speed and memory use) sets the ranking. 60.7 tok/s Runs great
LFM2.5 230MLiquid AI · 0.2B 10Basic quality (D) · 10 of 100: how capable the model is at F16, whatever it runs on. Match score 51/100 (quality, fit, speed and memory use) sets the ranking. 582 tok/s Runs great 10Basic quality (D) · 10 of 100: how capable the model is at F16, whatever it runs on. Match score 51/100 (quality, fit, speed and memory use) sets the ranking. 142 tok/s Runs great
Qwen2.5-Math 72B InstructAlibaba · 73B 40Fair quality (C) · 40 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 51/100 (quality, fit, speed and memory use) sets the ranking. 11.4 tok/s Runs welloffloads to RAM 40Fair quality (C) · 40 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 50/100 (quality, fit, speed and memory use) sets the ranking. 3.2 tok/s Slow
SmolLM2 360M InstructHugging Face · 0.4B 8Basic quality (D) · 8 of 100: how capable the model is at F16, whatever it runs on. Match score 50/100 (quality, fit, speed and memory use) sets the ranking. 372 tok/s Runs great 8Basic quality (D) · 8 of 100: how capable the model is at F16, whatever it runs on. Match score 50/100 (quality, fit, speed and memory use) sets the ranking. 91.0 tok/s Runs great
TinyLlama 1.1B ChatTinyLlama · 1.1B 6Basic quality (D) · 6 of 100: how capable the model is at F16, whatever it runs on. Match score 50/100 (quality, fit, speed and memory use) sets the ranking. 122 tok/s Runs great 6Basic quality (D) · 6 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 49/100 (quality, fit, speed and memory use) sets the ranking. 64.6 tok/s Runs great
Command R+ (08-2024) 104BCohere · 104B 49Fair quality (C) · 49 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 50/100 (quality, fit, speed and memory use) sets the ranking. 3.0 tok/s Slowheavy RAM offload 69Good quality (B) · 69 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 0.9 tok/s Very slow
DeepSeek Coder 1.3B InstructDeepSeek · 1.4B 12Basic quality (D) · 12 of 100: how capable the model is at F16, whatever it runs on. Match score 50/100 (quality, fit, speed and memory use) sets the ranking. 99.1 tok/s Runs great 12Basic quality (D) · 12 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 47/100 (quality, fit, speed and memory use) sets the ranking. 52.7 tok/s Runs great
LLM4Decompile 1.3B V2LLM4Binary · 1.4B 12Basic quality (D) · 12 of 100: how capable the model is at F16, whatever it runs on. Match score 50/100 (quality, fit, speed and memory use) sets the ranking. 99.1 tok/s Runs great 12Basic quality (D) · 12 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 47/100 (quality, fit, speed and memory use) sets the ranking. 52.7 tok/s Runs great
Qwen2 0.5B InstructAlibaba · 0.5B 6Basic quality (D) · 6 of 100: how capable the model is at F16, whatever it runs on. Match score 50/100 (quality, fit, speed and memory use) sets the ranking. 273 tok/s Runs great 6Basic quality (D) · 6 of 100: how capable the model is at F16, whatever it runs on. Match score 49/100 (quality, fit, speed and memory use) sets the ranking. 66.9 tok/s Runs great
ERNIE 4.5 0.3BBaidu · 0.4B 6Basic quality (D) · 6 of 100: how capable the model is at F16, whatever it runs on. Match score 49/100 (quality, fit, speed and memory use) sets the ranking. 372 tok/s Runs great 6Basic quality (D) · 6 of 100: how capable the model is at F16, whatever it runs on. Match score 49/100 (quality, fit, speed and memory use) sets the ranking. 91.0 tok/s Runs great
SmolVLM 256M InstructHugging Face · 0.3B 6Basic quality (D) · 6 of 100: how capable the model is at F16, whatever it runs on. Match score 49/100 (quality, fit, speed and memory use) sets the ranking. 514 tok/s Runs great 6Basic quality (D) · 6 of 100: how capable the model is at F16, whatever it runs on. Match score 49/100 (quality, fit, speed and memory use) sets the ranking. 126 tok/s Runs great
Gemma 3 270MGoogle · 0.3B 6Basic quality (D) · 6 of 100: how capable the model is at F16, whatever it runs on. Match score 49/100 (quality, fit, speed and memory use) sets the ranking. 495 tok/s Runs great 6Basic quality (D) · 6 of 100: how capable the model is at F16, whatever it runs on. Match score 49/100 (quality, fit, speed and memory use) sets the ranking. 121 tok/s Runs great
SmolLM 360M InstructHugging Face · 0.4B 5Basic quality (D) · 5 of 100: how capable the model is at F16, whatever it runs on. Match score 49/100 (quality, fit, speed and memory use) sets the ranking. 372 tok/s Runs great 5Basic quality (D) · 5 of 100: how capable the model is at F16, whatever it runs on. Match score 49/100 (quality, fit, speed and memory use) sets the ranking. 91.0 tok/s Runs great
surya ocr 2.5Datalab · 0.7B 15Basic quality (D) · 15 of 100: how capable the model is at F16, whatever it runs on. Match score 49/100 (quality, fit, speed and memory use) sets the ranking. 194 tok/s Runs great 15Basic quality (D) · 15 of 100: how capable the model is at F16, whatever it runs on. Match score 48/100 (quality, fit, speed and memory use) sets the ranking. 47.5 tok/s Runs great
Command A Reasoning (111B)Cohere · 111B 57Good quality (B) · 57 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 48/100 (quality, fit, speed and memory use) sets the ranking. 2.8 tok/s Very slowheavy RAM offload 82Great quality (A) · 82 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. 0.6 tok/s Very slow
SmolLM2 135M InstructHugging Face · 0.1B 4Basic quality (D) · 4 of 100: how capable the model is at F16, whatever it runs on. Match score 48/100 (quality, fit, speed and memory use) sets the ranking. 1,029 tok/s Runs great 4Basic quality (D) · 4 of 100: how capable the model is at F16, whatever it runs on. Match score 48/100 (quality, fit, speed and memory use) sets the ranking. 252 tok/s Runs great
Command R+ 104B (04-2024)Cohere · 104B 44Fair quality (C) · 44 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 48/100 (quality, fit, speed and memory use) sets the ranking. 3.0 tok/s Slowheavy RAM offload 63Good quality (B) · 63 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 0.9 tok/s Very slow
Falcon-H1 Tiny 90M InstructTII · 0.1B 3Basic quality (D) · 3 of 100: how capable the model is at F16, whatever it runs on. Match score 48/100 (quality, fit, speed and memory use) sets the ranking. 1,486 tok/s Runs great 3Basic quality (D) · 3 of 100: how capable the model is at F16, whatever it runs on. Match score 48/100 (quality, fit, speed and memory use) sets the ranking. 364 tok/s Runs great
SmolLM 135M InstructHugging Face · 0.1B 2Basic quality (D) · 2 of 100: how capable the model is at F16, whatever it runs on. Match score 47/100 (quality, fit, speed and memory use) sets the ranking. 1,029 tok/s Runs great 2Basic quality (D) · 2 of 100: how capable the model is at F16, whatever it runs on. Match score 47/100 (quality, fit, speed and memory use) sets the ranking. 252 tok/s Runs great
Qwen2.5 Coder 0.5B InstructAlibaba · 0.5B 8Basic quality (D) · 8 of 100: how capable the model is at F16, whatever it runs on. Match score 45/100 (quality, fit, speed and memory use) sets the ranking. 273 tok/s Runs great 8Basic quality (D) · 8 of 100: how capable the model is at F16, whatever it runs on. Match score 45/100 (quality, fit, speed and memory use) sets the ranking. 66.9 tok/s Runs great
Qwen1.5 110B ChatAlibaba · 111B 36Basic quality (D) · 36 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 44/100 (quality, fit, speed and memory use) sets the ranking. 2.5 tok/s Very slowheavy RAM offload 51Fair quality (C) · 51 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 0.9 tok/s Very slow
gpt-oss 120BOpenAI · 117B (5.1B active) 84Great quality (A) · 84 of 100: how capable the model is at MXFP4, whatever it runs on. Match score 11/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 84Great quality (A) · 84 of 100: how capable the model is at MXFP4, whatever it runs on. Match score 85/100 (quality, fit, speed and memory use) sets the ranking. 26.7 tok/s Runs well
Kimi K3.1 Pro (2.8T)Moonshot AI · 2,780B (104.0B active) 69Good quality (B) · 69 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 9/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 69Good quality (B) · 69 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 9/100 (quality, fit, speed and memory use) sets the ranking. — Won't run
DeepSeek V4.1 Flash 1017 (552B-A16B)DeepSeek · 763B (16.0B active) 68Good quality (B) · 68 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 9/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 82Great quality (A) · 82 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 88/100 (quality, fit, speed and memory use) sets the ranking. 10.9 tok/s Runs welltight fit
GLM-5.3 (753B-A40B)Zhipu AI · 753B (40.0B active) 68Good quality (B) · 68 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 9/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 82Great quality (A) · 82 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 86/100 (quality, fit, speed and memory use) sets the ranking. 4.4 tok/s Slowtight fit
Qwen3.8 Max (2.4T-A95B)Alibaba · 2,446B (95.0B active) 68Good quality (B) · 68 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 9/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 68Good quality (B) · 68 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 9/100 (quality, fit, speed and memory use) sets the ranking. — Won't run
Kimi K3 (2.8T)Moonshot AI · 2,780B (104.0B active) 68Good quality (B) · 68 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 9/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 68Good quality (B) · 68 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 9/100 (quality, fit, speed and memory use) sets the ranking. — Won't run
MiMo V2.6 Pro (1T-A42B)Xiaomi · 1,024B (42.0B active) 68Good quality (B) · 68 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 9/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 68Good quality (B) · 68 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. 5.4 tok/s Slowtight fit
Ling 3.1 Flash 1019 (560B-A25B)Ant Group (inclusionAI) · 572B (25.0B active) 68Good quality (B) · 68 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 9/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 92Excellent quality (A+) · 92 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 92/100 (quality, fit, speed and memory use) sets the ranking. 5.5 tok/s Slow
MiniMax M3.1 Flash (428B-A23B)MiniMax · 429B (23.0B active) 67Good quality (B) · 67 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 9/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 96Excellent quality (A+) · 96 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 93/100 (quality, fit, speed and memory use) sets the ranking. 4.2 tok/s Slow
Step 5 Preview (600B-A27B)StepFun · 604B (27.0B active) 67Good quality (B) · 67 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 9/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 94Excellent quality (A+) · 94 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 91/100 (quality, fit, speed and memory use) sets the ranking. 4.2 tok/s Slowtight fit
GLM-5.3 Flash (320B-A18B)Zhipu AI · 321B (18.0B active) 67Good quality (B) · 67 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 9/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 67Good quality (B) · 67 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking. 12.3 tok/s Runs well
MiMo V2.6 Flash (310B-A15B)Xiaomi · 311B (15.0B active) 67Good quality (B) · 67 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 9/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 81Great quality (A) · 81 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 83/100 (quality, fit, speed and memory use) sets the ranking. 11.6 tok/s Runs well
DeepSeek V4.1 Flash (552B-A16B)DeepSeek · 763B (16.0B active) 67Good quality (B) · 67 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 9/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 81Great quality (A) · 81 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 88/100 (quality, fit, speed and memory use) sets the ranking. 10.9 tok/s Runs welltight fit
Mistral Large 4 (1T-A52B)Mistral AI · 1,053B (52.0B active) 67Good quality (B) · 67 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 9/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 67Good quality (B) · 67 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. 4.4 tok/s Slowtight fit
Qwen3.8 Flash Next (125B-A6B)Alibaba · 180B (6.0B active) 67Good quality (B) · 67 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 9/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 91Excellent quality (A+) · 91 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 89/100 (quality, fit, speed and memory use) sets the ranking. 21.6 tok/s Runs well
GLM-5.2 (753B-A40B)Zhipu AI · 753B (40.0B active) 67Good quality (B) · 67 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 9/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 80Great quality (A) · 80 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 85/100 (quality, fit, speed and memory use) sets the ranking. 4.4 tok/s Slowtight fit
Hunyuan Hy4 Preview (780B)Tencent · 780B (55.0B active) 67Good quality (B) · 67 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 9/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 80Great quality (A) · 80 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 84/100 (quality, fit, speed and memory use) sets the ranking. 3.2 tok/s Slowtight fit
Muse Spark Open (816B-A42B)Meta · 822B (42.0B active) 67Good quality (B) · 67 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 9/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 80Great quality (A) · 80 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 85/100 (quality, fit, speed and memory use) sets the ranking. 4.2 tok/s Slowtight fit
DeepSeek V4 Pro (1.6T-A49B)DeepSeek · 1,651B (49.0B active) 67Good quality (B) · 67 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 9/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 67Good quality (B) · 67 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 9/100 (quality, fit, speed and memory use) sets the ranking. — Won't run
Kimi K2.6 (1T)Moonshot AI · 1,027B (32.0B active) 66Good quality (B) · 66 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 9/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 66Good quality (B) · 66 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. 7.0 tok/s Slowtight fit
Ling 3.1 Flash (560B-A25B)Ant Group (inclusionAI) · 572B (25.0B active) 66Good quality (B) · 66 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 9/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 90Excellent quality (A+) · 90 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 91/100 (quality, fit, speed and memory use) sets the ranking. 5.5 tok/s Slow
MiMo V2.5 Pro (1T-A42B)Xiaomi · 1,023B (42.0B active) 66Good quality (B) · 66 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 9/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 66Good quality (B) · 66 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. 5.4 tok/s Slowtight fit
Intern-S2 (397B-A17B)Shanghai AI Lab · 403B (17.0B active) 66Good quality (B) · 66 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 9/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 80Great quality (A) · 80 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 85/100 (quality, fit, speed and memory use) sets the ranking. 10.3 tok/s Runs well
GLM-5.1 (744B-A40B)Zhipu AI · 754B (40.0B active) 66Good quality (B) · 66 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 79Great quality (A) · 79 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 84/100 (quality, fit, speed and memory use) sets the ranking. 4.4 tok/s Slowtight fit
Nex N2 Pro (397B-A17B)Nex-AGI · 397B (17.0B active) 66Good quality (B) · 66 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 79Great quality (A) · 79 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 84/100 (quality, fit, speed and memory use) sets the ranking. 10.3 tok/s Runs well
DeepSeek V4 Pro (April 2026)DeepSeek · 1,599B (49.0B active) 66Good quality (B) · 66 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 66Good quality (B) · 66 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run
Naive N0.5 Flash (309B-A15.5B)Naive AI · 309B (15.5B active) 66Good quality (B) · 66 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 79Great quality (A) · 79 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 83/100 (quality, fit, speed and memory use) sets the ranking. 11.2 tok/s Runs well
Qwen3.5 397B-A17BAlibaba · 403B (17.0B active) 66Good quality (B) · 66 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 79Great quality (A) · 79 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 85/100 (quality, fit, speed and memory use) sets the ranking. 10.3 tok/s Runs well
MiniMax M3 (427B)MiniMax · 427B (25.0B active) 66Good quality (B) · 66 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 93Excellent quality (A+) · 93 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 91/100 (quality, fit, speed and memory use) sets the ranking. 3.8 tok/s Slow
GLM-5 (744B-A40B)Zhipu AI · 754B (40.0B active) 66Good quality (B) · 66 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 79Great quality (A) · 79 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 84/100 (quality, fit, speed and memory use) sets the ranking. 4.4 tok/s Slowtight fit
K2-Horizon (375B-A23B)Institute of Foundation Models · 379B (23.0B active) 66Good quality (B) · 66 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 93Excellent quality (A+) · 93 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 91/100 (quality, fit, speed and memory use) sets the ranking. 4.2 tok/s Slow
Hunyuan Hy3 (295B-A21B)Tencent · 299B (21.0B active) 66Good quality (B) · 66 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 66Good quality (B) · 66 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking. 10.6 tok/s Runs well
Beam 1.0 (501B-A23B)Reflection AI · 501B (23.4B active) 66Good quality (B) · 66 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 91Excellent quality (A+) · 91 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 91/100 (quality, fit, speed and memory use) sets the ranking. 4.8 tok/s Slow
Kimi K2.5 (1T)Moonshot AI · 1,027B (32.0B active) 66Good quality (B) · 66 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 66Good quality (B) · 66 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. 7.0 tok/s Slowtight fit
Kimi K2.7 Code (1T)Moonshot AI · 1,027B (32.0B active) 66Good quality (B) · 66 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 66Good quality (B) · 66 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. 7.0 tok/s Slowtight fit
DeepSeek V4 Flash (284B-A13B)DeepSeek · 304B (13.0B active) 65Good quality (B) · 65 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 89Excellent quality (A+) · 89 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 88/100 (quality, fit, speed and memory use) sets the ranking. 10.4 tok/s Runs well
MiMo V2.5 (310B-A15B)Xiaomi · 311B (15.0B active) 65Good quality (B) · 65 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 78Great quality (A) · 78 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 82/100 (quality, fit, speed and memory use) sets the ranking. 11.6 tok/s Runs well
Ornith 1.5 397B-A17BDeepReinforce · 403B (17.0B active) 65Good quality (B) · 65 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 78Great quality (A) · 78 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 84/100 (quality, fit, speed and memory use) sets the ranking. 10.3 tok/s Runs well
DeepSeek V3.2 (671B)DeepSeek · 685B (37.0B active) 65Good quality (B) · 65 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 89Excellent quality (A+) · 89 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 88/100 (quality, fit, speed and memory use) sets the ranking. 3.7 tok/s Slowtight fit
Inkling Small (276B-A12B)Thinking Machines Lab · 266B (12.0B active) 65Good quality (B) · 65 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 89Excellent quality (A+) · 89 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 87/100 (quality, fit, speed and memory use) sets the ranking. 11.2 tok/s Runs well
GLM-4.7 (355B-A32B)Zhipu AI · 358B (32.0B active) 65Good quality (B) · 65 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 90Excellent quality (A+) · 90 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 89/100 (quality, fit, speed and memory use) sets the ranking. 3.6 tok/s Slow
Inkling (975B-A41B)Thinking Machines Lab · 952B (41.0B active) 65Good quality (B) · 65 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 65Good quality (B) · 65 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. 5.5 tok/s Slowtight fit
MiniMax M2.7 (230B-A10B)MiniMax · 229B (10.0B active) 65Good quality (B) · 65 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 90Excellent quality (A+) · 90 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 88/100 (quality, fit, speed and memory use) sets the ranking. 11.1 tok/s Runs well
Step 3.7 Flash (198B-A11B)StepFun · 201B (11.0B active) 65Good quality (B) · 65 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 90Excellent quality (A+) · 90 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 87/100 (quality, fit, speed and memory use) sets the ranking. 10.1 tok/s Runs well
Nemotron 3 Ultra 550B-A55BNVIDIA · 561B (55.0B active) 65Good quality (B) · 65 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 78Great quality (A) · 78 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 84/100 (quality, fit, speed and memory use) sets the ranking. 3.2 tok/s Slow
DeepSeek R1 0528 (671B)DeepSeek · 685B (37.0B active) 65Good quality (B) · 65 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 88Excellent quality (A+) · 88 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 83/100 (quality, fit, speed and memory use) sets the ranking. 3.7 tok/s Slowtight fit
Kimi K2 Thinking (1T)Moonshot AI · 1,026B (32.0B active) 65Good quality (B) · 65 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 65Good quality (B) · 65 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. 7.0 tok/s Slowtight fit
DeepSeek V3.2 ExpDeepSeek · 685B (37.0B active) 64Good quality (B) · 64 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 87Excellent quality (A+) · 87 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 88/100 (quality, fit, speed and memory use) sets the ranking. 3.7 tok/s Slowtight fit
Ling 2.6 1TAnt Group (inclusionAI) · 1,025B (63.0B active) 64Good quality (B) · 64 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 64Good quality (B) · 64 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking. 3.6 tok/s Slowtight fit
MiniMax M2.5 (230B-A10B)MiniMax · 229B (10.0B active) 64Good quality (B) · 64 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 89Excellent quality (A+) · 89 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 88/100 (quality, fit, speed and memory use) sets the ranking. 11.1 tok/s Runs well
DeepSeek R1 (671B)DeepSeek · 671B (37.0B active) 64Good quality (B) · 64 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 87Excellent quality (A+) · 87 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 87/100 (quality, fit, speed and memory use) sets the ranking. 3.7 tok/s Slowtight fit
DeepSeek V4 Flash (April 2026)DeepSeek · 291B (13.0B active) 64Good quality (B) · 64 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 87Excellent quality (A+) · 87 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 82/100 (quality, fit, speed and memory use) sets the ranking. 10.4 tok/s Runs well
Step 3.5 Flash (196B-A11B)StepFun · 199B (11.0B active) 64Good quality (B) · 64 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 89Excellent quality (A+) · 89 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 86/100 (quality, fit, speed and memory use) sets the ranking. 10.1 tok/s Runs well
DeepSeek V3.1 TerminusDeepSeek · 685B (37.0B active) 64Good quality (B) · 64 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 86Excellent quality (A+) · 86 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 87/100 (quality, fit, speed and memory use) sets the ranking. 3.7 tok/s Slowtight fit
DeepSeek V3.1 (671B)DeepSeek · 685B (37.0B active) 63Good quality (B) · 63 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 86Excellent quality (A+) · 86 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 87/100 (quality, fit, speed and memory use) sets the ranking. 3.7 tok/s Slowtight fit
Kimi K2 Instruct 0905 (1T)Moonshot AI · 1,026B (32.0B active) 63Good quality (B) · 63 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 63Good quality (B) · 63 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 78/100 (quality, fit, speed and memory use) sets the ranking. 7.0 tok/s Slowtight fit
MiniMax M2.1 (230B-A10B)MiniMax · 229B (10.0B active) 63Good quality (B) · 63 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 88Excellent quality (A+) · 88 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 87/100 (quality, fit, speed and memory use) sets the ranking. 11.1 tok/s Runs well
Qwen3 235B-A22B Thinking 2507Alibaba · 235B (22.0B active) 63Good quality (B) · 63 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 63Good quality (B) · 63 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 10.1 tok/s Runs well
Qwen3 Coder 480B-A35BAlibaba · 480B (35.0B active) 63Good quality (B) · 63 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 88Excellent quality (A+) · 88 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 84/100 (quality, fit, speed and memory use) sets the ranking. 3.3 tok/s Slow
MiMo V2 Flash (309B-A15B)Xiaomi · 310B (15.0B active) 63Good quality (B) · 63 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 76Great quality (A) · 76 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking. 11.6 tok/s Runs well
Ling 1TAnt Group (inclusionAI) · 1,000B (50.0B active) 63Good quality (B) · 63 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 63Good quality (B) · 63 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking. 4.5 tok/s Slowtight fit
Mistral Large 3 (675B)Mistral AI · 675B (41.0B active) 63Good quality (B) · 63 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 85Excellent quality (A+) · 85 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 86/100 (quality, fit, speed and memory use) sets the ranking. 3.4 tok/s Slowtight fit
Qwen3 235B-A22B Instruct 2507Alibaba · 235B (22.0B active) 63Good quality (B) · 63 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 63Good quality (B) · 63 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. 10.1 tok/s Runs well
Solar Open 2 250BUpstage · 250B (15.0B active) 63Good quality (B) · 63 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 75Great quality (A) · 75 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. 11.6 tok/s Runs well
Kimi K2 Instruct (July 2025)Moonshot AI · 1,026B (32.0B active) 62Good quality (B) · 62 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 62Good quality (B) · 62 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 78/100 (quality, fit, speed and memory use) sets the ranking. 7.0 tok/s Slowtight fit
MiniMax M2 (230B-A10B)MiniMax · 229B (10.0B active) 62Good quality (B) · 62 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 86Excellent quality (A+) · 86 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 86/100 (quality, fit, speed and memory use) sets the ranking. 11.1 tok/s Runs well
Qwen3-VL 235B-A22B InstructAlibaba · 236B (22.0B active) 62Good quality (B) · 62 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 62Good quality (B) · 62 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. 10.1 tok/s Runs well
Command A+ (218B-A25B)Cohere · 219B (25.0B active) 61Good quality (B) · 61 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 87Excellent quality (A+) · 87 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 84/100 (quality, fit, speed and memory use) sets the ranking. 3.8 tok/s Slow
K-EXAONE 236B-A23BLG AI Research · 237B (23.0B active) 61Good quality (B) · 61 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 87Excellent quality (A+) · 87 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 85/100 (quality, fit, speed and memory use) sets the ranking. 4.2 tok/s Slow
LongCat Flash Chat (560B)Meituan · 562B (27.0B active) 61Good quality (B) · 61 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 83Great quality (A) · 83 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 87/100 (quality, fit, speed and memory use) sets the ranking. 5.1 tok/s Slow
Qwen3 235B-A22BAlibaba · 235B (22.0B active) 61Good quality (B) · 61 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 61Good quality (B) · 61 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. 10.1 tok/s Runs well
Devstral 2 (123B)Mistral AI · 125B 60Good quality (B) · 60 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 86Excellent quality (A+) · 86 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. 0.6 tok/s Very slow
InternVL3.5 241B-A28BShanghai AI Lab · 241B (28.0B active) 60Good quality (B) · 60 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 85Excellent quality (A+) · 85 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 84/100 (quality, fit, speed and memory use) sets the ranking. 3.4 tok/s Slow
Llama 3.1 Nemotron Ultra 253B v1NVIDIA · 253B 60Good quality (B) · 60 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 8/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 86Excellent quality (A+) · 86 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 84/100 (quality, fit, speed and memory use) sets the ranking. 0.3 tok/s Very slow
Baichuan M3 235B-A22BBaichuan · 235B (22.0B active) 58Good quality (B) · 58 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 7/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 58Good quality (B) · 58 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking. 10.1 tok/s Runs well
GLM-4.6 (355B-A32B)Zhipu AI · 357B (32.0B active) 58Good quality (B) · 58 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 7/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 81Great quality (A) · 81 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 85/100 (quality, fit, speed and memory use) sets the ranking. 3.6 tok/s Slow
DeepSeek V3 0324 (671B)DeepSeek · 685B (37.0B active) 58Good quality (B) · 58 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 7/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 78Great quality (A) · 78 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 84/100 (quality, fit, speed and memory use) sets the ranking. 3.7 tok/s Slowtight fit
MiniMax-M1 80KMiniMax · 456B (45.9B active) 58Good quality (B) · 58 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 7/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 78Great quality (A) · 78 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 84/100 (quality, fit, speed and memory use) sets the ranking. 3.0 tok/s Slow
GLM-4.5 (355B-A32B)Zhipu AI · 358B (32.0B active) 57Good quality (B) · 57 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 7/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 79Great quality (A) · 79 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 84/100 (quality, fit, speed and memory use) sets the ranking. 3.6 tok/s Slow
Ling 3.0 Flash (127B-A5B)Ant Group (inclusionAI) · 127B (5.1B active) 57Good quality (B) · 57 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 7/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 77Great quality (A) · 77 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 82/100 (quality, fit, speed and memory use) sets the ranking. 25.1 tok/s Runs well
MiniMax-M1 40KMiniMax · 456B (45.9B active) 57Good quality (B) · 57 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 7/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 77Great quality (A) · 77 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 84/100 (quality, fit, speed and memory use) sets the ranking. 3.0 tok/s Slow
Mistral Medium 3.5 (128B)Mistral AI · 128B 56Good quality (B) · 56 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 7/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 80Great quality (A) · 80 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking. 0.6 tok/s Very slow
Trinity Large Thinking (400B-A13B)Arcee AI · 399B (13.0B active) 56Good quality (B) · 56 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 7/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 76Great quality (A) · 76 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. 10.4 tok/s Runs well
ERNIE 4.5 300B-A47BBaidu · 300B (47.0B active) 56Good quality (B) · 56 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 7/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 67Good quality (B) · 67 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. 3.8 tok/s Slow
Hermes 4 405BNous Research · 406B 56Good quality (B) · 56 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 7/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 79Great quality (A) · 79 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking. 0.2 tok/s Very slow
Mistral Large 2.1 (2411)Mistral AI · 123B 56Good quality (B) · 56 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 7/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 80Great quality (A) · 80 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking. 0.6 tok/s Very slow
Mistral Large 2 (123B)Mistral AI · 123B 54Fair quality (C) · 54 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 7/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 78Great quality (A) · 78 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking. 0.6 tok/s Very slow
NorthCUA Probeipo-ai · 397B (17.0B active) 54Fair quality (C) · 54 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 7/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 65Good quality (B) · 65 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 78/100 (quality, fit, speed and memory use) sets the ranking. 10.3 tok/s Runs well
Pixtral Large 124B (2411)Mistral AI · 124B 54Fair quality (C) · 54 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 7/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 78Great quality (A) · 78 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking. 0.6 tok/s Very slow
dots.llm1 Instructrednote · 143B (14.0B active) 53Fair quality (C) · 53 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 7/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 63Good quality (B) · 63 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. 12.4 tok/s Runs well
MiniMax-Text-01MiniMax · 456B (45.9B active) 53Fair quality (C) · 53 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 7/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 72Great quality (A) · 72 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking. 3.0 tok/s Slow
Tülu 3 405BAi2 · 406B 53Fair quality (C) · 53 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 7/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 75Great quality (A) · 75 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. 0.2 tok/s Very slow
DeepSeek V3 (671B)DeepSeek · 671B (37.0B active) 52Fair quality (C) · 52 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 7/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 71Great quality (A) · 71 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. 3.7 tok/s Slowtight fit
DeepSeek Coder V2 (236B)DeepSeek · 236B (21.0B active) 51Fair quality (C) · 51 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 6/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 51Fair quality (C) · 51 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. 10.6 tok/s Runs well
DeepSeek Coder V2 Instruct (0614)DeepSeek · 236B (21.0B active) 50Fair quality (C) · 50 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 6/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 50Fair quality (C) · 50 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 10.6 tok/s Runs well
DeepSeek V2.5 1210DeepSeek · 236B (21.0B active) 50Fair quality (C) · 50 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 6/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 50Fair quality (C) · 50 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 10.6 tok/s Runs well
Llama 3.1 405B InstructMeta · 406B 50Fair quality (C) · 50 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 6/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 71Great quality (A) · 71 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 78/100 (quality, fit, speed and memory use) sets the ranking. 0.2 tok/s Very slow
Llama 4 Maverick (17B x 128E)Meta · 402B (17.0B active) 50Fair quality (C) · 50 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 6/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 60Good quality (B) · 60 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking. 10.3 tok/s Runs well
DeepSeek V2.5DeepSeek · 236B (21.0B active) 49Fair quality (C) · 49 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 6/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 49Fair quality (C) · 49 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 10.6 tok/s Runs well
Jamba 1.5 LargeAI21 Labs · 399B (94.0B active) 49Fair quality (C) · 49 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 6/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 69Good quality (B) · 69 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 78/100 (quality, fit, speed and memory use) sets the ranking. 1.0 tok/s Very slow
Ling 3.0 flash SanteAnt Group (inclusionAI) · 127B (5.1B active) 47Fair quality (C) · 47 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 6/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 64Good quality (B) · 64 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking. 25.1 tok/s Runs well
DeepSeek V2 ChatDeepSeek · 236B (21.0B active) 43Fair quality (C) · 43 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 5/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 43Fair quality (C) · 43 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 10.6 tok/s Runs well
Mixtral 8x22B InstructMistral AI · 141B (39.0B active) 38Basic quality (D) · 38 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 5/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 52Fair quality (C) · 52 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 3.5 tok/s Slow
Grok-1 (314B)xAI · 314B (86.0B active) 35Basic quality (D) · 35 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 4/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 49Fair quality (C) · 49 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 1.1 tok/s Very slow
Snowflake Arctic InstructSnowflake · 479B (17.0B active) 35Basic quality (D) · 35 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 4/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 42Fair quality (C) · 42 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 10.3 tok/s Runs well
Falcon 180B ChatTII · 180B 23Basic quality (D) · 23 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 3/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 33Basic quality (D) · 33 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 55/100 (quality, fit, speed and memory use) sets the ranking. 0.5 tok/s Very slow

Image models

59 of 60 image models run on the Intel Arc Pro B70. 32.0 GB usable for models.

60 of 60 image models run on the AMD EPYC 7002 / 7003 series NAS (512 GB). 471 GB usable for models.

Model QualityArc Pro B70 SpeedArc Pro B70 RunsArc Pro B70 QualityEPYC 7002 / 7003 series NAS SpeedEPYC 7002 / 7003 series NAS RunsEPYC 7002 / 7003 series NAS
Krea 2 TurboKrea · 12.8B 90Excellent quality (A+) · 90 of 100: how capable the model is at FP8, whatever it runs on. Match score 93/100 (quality, fit, speed and memory use) sets the ranking. 13 s Runs great 92Excellent quality (A+) · 92 of 100: how capable the model is at FP16, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. 2.9 min Very slow
ERNIE-Image TurboBaidu · 8.0B 85Excellent quality (A+) · 85 of 100: how capable the model is at FP16, whatever it runs on. Match score 92/100 (quality, fit, speed and memory use) sets the ranking. 10 s Runs great 85Excellent quality (A+) · 85 of 100: how capable the model is at FP16, whatever it runs on. Match score 68/100 (quality, fit, speed and memory use) sets the ranking. 2.2 min Slow
Z-Image TurboAlibaba (Tongyi) · 6.2B 84Great quality (A) · 84 of 100: how capable the model is at FP16, whatever it runs on. Match score 92/100 (quality, fit, speed and memory use) sets the ranking. 8.2 s Runs great 84Great quality (A) · 84 of 100: how capable the model is at FP16, whatever it runs on. Match score 68/100 (quality, fit, speed and memory use) sets the ranking. 1.7 min Slow
FLUX.2 [klein] 9BBlack Forest Labs · 9.1B 83Great quality (A) · 83 of 100: how capable the model is at Q8 GGUF, whatever it runs on. Match score 92/100 (quality, fit, speed and memory use) sets the ranking. 6.8 s Runs great 84Great quality (A) · 84 of 100: how capable the model is at FP16, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking. 74 s Slow
FLUX 3 Image KleinBlack Forest Labs · 5.6B 91Excellent quality (A+) · 91 of 100: how capable the model is at FP16, whatever it runs on. Match score 91/100 (quality, fit, speed and memory use) sets the ranking. 19 s Runs well 91Excellent quality (A+) · 91 of 100: how capable the model is at FP16, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking. 3.9 min Very slow
FLUX.2 [klein] 4BBlack Forest Labs · 3.9B 78Great quality (A) · 78 of 100: how capable the model is at FP16, whatever it runs on. Match score 88/100 (quality, fit, speed and memory use) sets the ranking. 3.3 s Runs great 76Great quality (A) · 76 of 100: how capable the model is at FP8, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 37 s Runs well
FLUX.1 [schnell]Black Forest Labs · 11.9B 69Good quality (B) · 69 of 100: how capable the model is at Q8 GGUF, whatever it runs on. Match score 85/100 (quality, fit, speed and memory use) sets the ranking. 7.5 s Runs great 70Great quality (A) · 70 of 100: how capable the model is at FP16, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 82 s Slow
Fibo 2.0 DevBria AI · 8.1B 93Excellent quality (A+) · 93 of 100: how capable the model is at FP16, whatever it runs on. Match score 85/100 (quality, fit, speed and memory use) sets the ranking. 73 s Slowtight fit 93Excellent quality (A+) · 93 of 100: how capable the model is at FP16, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 15.4 min Very slow
Qwen-Image 2.1 TurboAlibaba · 7.1B 87Excellent quality (A+) · 87 of 100: how capable the model is at FP8, whatever it runs on. Match score 85/100 (quality, fit, speed and memory use) sets the ranking. 34 s Runs welltight fit 89Excellent quality (A+) · 89 of 100: how capable the model is at FP16, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 7.9 min Very slow
Ideogram 4Ideogram · 18.6B 90Excellent quality (A+) · 90 of 100: how capable the model is at FP8, whatever it runs on. Match score 84/100 (quality, fit, speed and memory use) sets the ranking. 73 s Slowtight fit 92Excellent quality (A+) · 92 of 100: how capable the model is at FP16, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking. 17.1 min Very slow
FLUX.1 Krea [dev]Black Forest Labs · 11.9B 80Great quality (A) · 80 of 100: how capable the model is at FP8, whatever it runs on. Match score 83/100 (quality, fit, speed and memory use) sets the ranking. 35 s Runs well 82Great quality (A) · 82 of 100: how capable the model is at FP16, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 8.2 min Very slow
Z-ImageAlibaba (Tongyi) · 6.2B 82Great quality (A) · 82 of 100: how capable the model is at FP16, whatever it runs on. Match score 83/100 (quality, fit, speed and memory use) sets the ranking. 55 s Slow 82Great quality (A) · 82 of 100: how capable the model is at FP16, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 11.6 min Very slow
ERNIE-ImageBaidu · 8.0B 86Excellent quality (A+) · 86 of 100: how capable the model is at FP8, whatever it runs on. Match score 82/100 (quality, fit, speed and memory use) sets the ranking. 65 s Slow 88Excellent quality (A+) · 88 of 100: how capable the model is at FP16, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 15.1 min Very slow
FLUX.1 [dev]Black Forest Labs · 11.9B 78Great quality (A) · 78 of 100: how capable the model is at FP8, whatever it runs on. Match score 82/100 (quality, fit, speed and memory use) sets the ranking. 35 s Runs well 80Great quality (A) · 80 of 100: how capable the model is at FP16, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 8.2 min Very slow
FLUX.1 Kontext [dev]Black Forest Labs · 11.9B 80Great quality (A) · 80 of 100: how capable the model is at FP8, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking. 65 s Slow 82Great quality (A) · 82 of 100: how capable the model is at FP16, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 15.1 min Very slow
FLUX.1 Fill [dev]Black Forest Labs · 11.9B 74Great quality (A) · 74 of 100: how capable the model is at FP8, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. 37 s Runs well 76Great quality (A) · 76 of 100: how capable the model is at FP16, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 8.6 min Very slow
Qwen-Image 2512Alibaba · 20.4B 83Great quality (A) · 83 of 100: how capable the model is at Q4 GGUF, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. 5.9 min Very slow 90Excellent quality (A+) · 90 of 100: how capable the model is at FP16, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 61.6 min Very slow
Ming-Image 0.1 DesignAnt Group (inclusionAI) · 6.2B 86Excellent quality (A+) · 86 of 100: how capable the model is at FP16, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. 56 s Slowencoder in RAM 86Excellent quality (A+) · 86 of 100: how capable the model is at FP16, whatever it runs on. Match score 68/100 (quality, fit, speed and memory use) sets the ranking. 11.2 min Very slow
MOSS-Image SFTOpenMOSS · 5.1B 80Great quality (A) · 80 of 100: how capable the model is at FP8, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. 41 s Slow 82Great quality (A) · 82 of 100: how capable the model is at FP16, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 9.6 min Very slow
Stable Diffusion 3.5 Large TurboStability AI · 8.1B 61Good quality (B) · 61 of 100: how capable the model is at Q8 GGUF, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. 6.0 s Runs great 62Good quality (B) · 62 of 100: how capable the model is at FP16, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 66 s Slow
HiDream-I1 FastHiDream.ai · 17.1B 70Great quality (A) · 70 of 100: how capable the model is at NF4, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. 36 s Runs well 76Great quality (A) · 76 of 100: how capable the model is at FP16, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 6.8 min Very slow
HiDream-I1 DevHiDream.ai · 17.1B 74Great quality (A) · 74 of 100: how capable the model is at NF4, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. 61 s Slow 80Great quality (A) · 80 of 100: how capable the model is at FP16, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 11.6 min Very slow
FireRed-Image-Edit 1.1rednote (FireRed Team) · 20.4B 80Great quality (A) · 80 of 100: how capable the model is at Q4 GGUF, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. 5.2 min Very slow 87Excellent quality (A+) · 87 of 100: how capable the model is at FP16, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 54.8 min Very slow
HiDream-I1 FullHiDream.ai · 17.1B 78Great quality (A) · 78 of 100: how capable the model is at NF4, whatever it runs on. Match score 78/100 (quality, fit, speed and memory use) sets the ranking. 2.7 min Very slow 84Great quality (A) · 84 of 100: how capable the model is at FP16, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 30.8 min Very slow
Qwen-Image-Edit 2511Alibaba · 20.4B 79Great quality (A) · 79 of 100: how capable the model is at Q4 GGUF, whatever it runs on. Match score 78/100 (quality, fit, speed and memory use) sets the ranking. 5.2 min Very slow 86Excellent quality (A+) · 86 of 100: how capable the model is at FP16, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 54.8 min Very slow
FLUX.2 [dev]Black Forest Labs · 32.2B 85Excellent quality (A+) · 85 of 100: how capable the model is at Q4 GGUF, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking. 2.5 min Very slowencoder in RAM 92Excellent quality (A+) · 92 of 100: how capable the model is at FP16, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. 24.7 min Very slow
Chroma1-HDLodestone Rock · 8.9B 76Great quality (A) · 76 of 100: how capable the model is at FP8, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking. 59 s Slow 78Great quality (A) · 78 of 100: how capable the model is at FP16, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 13.7 min Very slow
Qwen-ImageAlibaba · 20.4B 86Excellent quality (A+) · 86 of 100: how capable the model is at FP8, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking. 4.6 min Very slowencoder in RAM 88Excellent quality (A+) · 88 of 100: how capable the model is at FP16, whatever it runs on. Match score 68/100 (quality, fit, speed and memory use) sets the ranking. 61.6 min Very slow
Iris-3BSperidlabs · 3.0B 72Great quality (A) · 72 of 100: how capable the model is at FP16, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking. 52 s Slow 72Great quality (A) · 72 of 100: how capable the model is at FP16, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 11.0 min Very slow
Stable Diffusion 3 MediumStability AI · 2.0B 52Fair quality (C) · 52 of 100: how capable the model is at FP16, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking. 10 s Runs great 52Fair quality (C) · 52 of 100: how capable the model is at FP16, whatever it runs on. Match score 53/100 (quality, fit, speed and memory use) sets the ranking. 2.1 min Slow
CogView4 6BZhipu AI · 6.4B 68Good quality (B) · 68 of 100: how capable the model is at FP8, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking. 53 s Slow 70Great quality (A) · 70 of 100: how capable the model is at FP16, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 12.3 min Very slow
Qwen-Image 2.1Alibaba · 7.1B 86Excellent quality (A+) · 86 of 100: how capable the model is at Q4 GGUF, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. 4.1 min Very slowtight fit · encoder in RAM 93Excellent quality (A+) · 93 of 100: how capable the model is at FP16, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking. 41.1 min Very slow
OmniGen2BAAI (VectorSpaceLab) · 4.0B 66Good quality (B) · 66 of 100: how capable the model is at FP16, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. 52 s Slow 66Good quality (B) · 66 of 100: how capable the model is at FP16, whatever it runs on. Match score 57/100 (quality, fit, speed and memory use) sets the ranking. 11.0 min Very slow
Stable Diffusion 3.5 LargeStability AI · 8.1B 64Good quality (B) · 64 of 100: how capable the model is at FP8, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. 32 s Runs well 66Good quality (B) · 66 of 100: how capable the model is at FP16, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. 7.5 min Very slow
KolorsKuaishou · 2.6B 58Good quality (B) · 58 of 100: how capable the model is at FP8, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. 12 s Runs great 60Good quality (B) · 60 of 100: how capable the model is at FP16, whatever it runs on. Match score 56/100 (quality, fit, speed and memory use) sets the ranking. 2.7 min Very slow
DreamShaper XL v2 TurboLykon · 2.6B 58Good quality (B) · 58 of 100: how capable the model is at FP16, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. 4.2 s Runs great 58Good quality (B) · 58 of 100: how capable the model is at FP16, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. 53 s Slow
Lumina-Image 2.0Shanghai AI Lab · 2.6B 62Good quality (B) · 62 of 100: how capable the model is at FP16, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. 23 s Runs well 62Good quality (B) · 62 of 100: how capable the model is at FP16, whatever it runs on. Match score 56/100 (quality, fit, speed and memory use) sets the ranking. 4.8 min Very slow
PixArt-Σ XL 1024PixArt · 0.6B 50Fair quality (C) · 50 of 100: how capable the model is at FP16, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. 4.9 s Runs great 50Fair quality (C) · 50 of 100: how capable the model is at FP16, whatever it runs on. Match score 54/100 (quality, fit, speed and memory use) sets the ranking. 62 s Slow
AuraFlow v0.3fal · 6.9B 58Good quality (B) · 58 of 100: how capable the model is at FP16, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. 46 s Slow 58Good quality (B) · 58 of 100: how capable the model is at FP16, whatever it runs on. Match score 53/100 (quality, fit, speed and memory use) sets the ranking. 9.6 min Very slow
Cosmos3-Super Text2Image (64B)NVIDIA · 64.0B 90Excellent quality (A+) · 90 of 100: how capable the model is at FP8, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. 2.5 min Very slowCPU offload 92Excellent quality (A+) · 92 of 100: how capable the model is at FP16, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. 27.4 min Very slow
SDXL-Lightning (4-step)ByteDance · 2.6B 54Fair quality (C) · 54 of 100: how capable the model is at FP16, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking. 2.3 s Runs great 54Fair quality (C) · 54 of 100: how capable the model is at FP16, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. 29 s Runs well
Kandinsky 3Sber AI · 3.0B 48Fair quality (C) · 48 of 100: how capable the model is at FP16, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking. 29 s Runs well 48Fair quality (C) · 48 of 100: how capable the model is at FP16, whatever it runs on. Match score 50/100 (quality, fit, speed and memory use) sets the ranking. 6.2 min Very slow
Juggernaut XL v9RunDiffusion · 2.6B 60Good quality (B) · 60 of 100: how capable the model is at FP8, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 13 s Runs great 62Good quality (B) · 62 of 100: how capable the model is at FP16, whatever it runs on. Match score 56/100 (quality, fit, speed and memory use) sets the ranking. 3.1 min Very slow
RealVisXL V5.0SG161222 · 2.6B 60Good quality (B) · 60 of 100: how capable the model is at FP8, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 13 s Runs great 62Good quality (B) · 62 of 100: how capable the model is at FP16, whatever it runs on. Match score 56/100 (quality, fit, speed and memory use) sets the ranking. 3.1 min Very slow
Stable Diffusion 3.5 MediumStability AI · 2.5B 56Good quality (B) · 56 of 100: how capable the model is at FP8, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 18 s Runs well 58Good quality (B) · 58 of 100: how capable the model is at FP16, whatever it runs on. Match score 55/100 (quality, fit, speed and memory use) sets the ranking. 4.1 min Very slow
Animagine XL 4.0Cagliostro Lab · 2.6B 58Good quality (B) · 58 of 100: how capable the model is at FP8, whatever it runs on. Match score 68/100 (quality, fit, speed and memory use) sets the ranking. 12 s Runs great 60Good quality (B) · 60 of 100: how capable the model is at FP16, whatever it runs on. Match score 56/100 (quality, fit, speed and memory use) sets the ranking. 2.9 min Very slow
Illustrious XL v0.1OnomaAI · 2.6B 58Good quality (B) · 58 of 100: how capable the model is at FP8, whatever it runs on. Match score 68/100 (quality, fit, speed and memory use) sets the ranking. 12 s Runs great 60Good quality (B) · 60 of 100: how capable the model is at FP16, whatever it runs on. Match score 56/100 (quality, fit, speed and memory use) sets the ranking. 2.9 min Very slow
Pony Diffusion V6 XLPurpleSmartAI · 2.6B 56Good quality (B) · 56 of 100: how capable the model is at FP8, whatever it runs on. Match score 68/100 (quality, fit, speed and memory use) sets the ranking. 11 s Runs great 58Good quality (B) · 58 of 100: how capable the model is at FP16, whatever it runs on. Match score 55/100 (quality, fit, speed and memory use) sets the ranking. 2.6 min Very slow
Playground v2.5Playground AI · 2.6B 60Good quality (B) · 60 of 100: how capable the model is at FP16, whatever it runs on. Match score 68/100 (quality, fit, speed and memory use) sets the ranking. 24 s Runs well 60Good quality (B) · 60 of 100: how capable the model is at FP16, whatever it runs on. Match score 55/100 (quality, fit, speed and memory use) sets the ranking. 5.1 min Very slow
Animagine XL 3.1Cagliostro Lab · 2.6B 54Fair quality (C) · 54 of 100: how capable the model is at FP8, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 12 s Runs great 56Good quality (B) · 56 of 100: how capable the model is at FP16, whatever it runs on. Match score 54/100 (quality, fit, speed and memory use) sets the ranking. 2.9 min Very slow
Stable Diffusion XL 1.0Stability AI · 2.6B 53Fair quality (C) · 53 of 100: how capable the model is at FP8, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 13 s Runs great 55Good quality (B) · 55 of 100: how capable the model is at FP16, whatever it runs on. Match score 53/100 (quality, fit, speed and memory use) sets the ranking. 3.1 min Very slow
Stable CascadeStability AI · 5.2B 50Fair quality (C) · 50 of 100: how capable the model is at FP8, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 15 s Runs well 52Fair quality (C) · 52 of 100: how capable the model is at FP16, whatever it runs on. Match score 52/100 (quality, fit, speed and memory use) sets the ranking. 3.4 min Very slow
Hunyuan-DiT v1.2Tencent · 1.5B 52Fair quality (C) · 52 of 100: how capable the model is at FP16, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 20 s Runs well 52Fair quality (C) · 52 of 100: how capable the model is at FP16, whatever it runs on. Match score 51/100 (quality, fit, speed and memory use) sets the ranking. 4.1 min Very slow
SDXL TurboStability AI · 2.6B 40Fair quality (C) · 40 of 100: how capable the model is at FP16, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 0.5 s Runs great 40Fair quality (C) · 40 of 100: how capable the model is at FP16, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. 6.2 s Runs great
Realistic Vision V5.1SG161222 · 0.9B 42Fair quality (C) · 42 of 100: how capable the model is at FP16, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 3.9 s Runs great 42Fair quality (C) · 42 of 100: how capable the model is at FP16, whatever it runs on. Match score 50/100 (quality, fit, speed and memory use) sets the ranking. 49 s Slow
DreamShaper 8Lykon · 0.9B 40Fair quality (C) · 40 of 100: how capable the model is at FP16, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 3.9 s Runs great 40Fair quality (C) · 40 of 100: how capable the model is at FP16, whatever it runs on. Match score 49/100 (quality, fit, speed and memory use) sets the ranking. 49 s Slow
Stable Diffusion 1.5Stability AI / Runway · 0.9B 35Basic quality (D) · 35 of 100: how capable the model is at FP16, whatever it runs on. Match score 57/100 (quality, fit, speed and memory use) sets the ranking. 3.9 s Runs great 35Basic quality (D) · 35 of 100: how capable the model is at FP16, whatever it runs on. Match score 47/100 (quality, fit, speed and memory use) sets the ranking. 49 s Slow
LCM DreamShaper v7Tsinghua (LCM) · 0.9B 30Basic quality (D) · 30 of 100: how capable the model is at FP16, whatever it runs on. Match score 55/100 (quality, fit, speed and memory use) sets the ranking. 0.8 s Runs great 30Basic quality (D) · 30 of 100: how capable the model is at FP16, whatever it runs on. Match score 53/100 (quality, fit, speed and memory use) sets the ranking. 10 s Runs great
SD TurboStability AI · 0.9B 28Basic quality (D) · 28 of 100: how capable the model is at FP16, whatever it runs on. Match score 54/100 (quality, fit, speed and memory use) sets the ranking. 0.3 s Runs great 28Basic quality (D) · 28 of 100: how capable the model is at FP16, whatever it runs on. Match score 52/100 (quality, fit, speed and memory use) sets the ranking. 3.3 s Runs great
HunyuanImage 3.0 (80B-A13B)Tencent · 83.0B 88Excellent quality (A+) · 88 of 100: how capable the model is at FP8, whatever it runs on. Match score 11/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 90Excellent quality (A+) · 90 of 100: how capable the model is at FP16, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. 41.1 min Very slow

Video models

40 of 43 video models run on the Intel Arc Pro B70. 32.0 GB usable for models.

43 of 43 video models run on the AMD EPYC 7002 / 7003 series NAS (512 GB). 471 GB usable for models.

Model QualityArc Pro B70 SpeedArc Pro B70 RunsArc Pro B70 QualityEPYC 7002 / 7003 series NAS SpeedEPYC 7002 / 7003 series NAS RunsEPYC 7002 / 7003 series NAS
LTX-Video 13B 0.9.8 (distilled)Lightricks · 13.0B 66Good quality (B) · 66 of 100: how capable the model is at FP8, whatever it runs on. Match score 82/100 (quality, fit, speed and memory use) sets the ranking. 2.9 min Runs welltight fit 66Good quality (B) · 66 of 100: how capable the model is at FP8, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 37.0 min Slow
LTX-2.5 (22B, video + audio)Lightricks · 21.0B 81Great quality (A) · 81 of 100: how capable the model is at Q4 GGUF, whatever it runs on. Match score 82/100 (quality, fit, speed and memory use) sets the ranking. 6.2 min Runs wellencoder in RAM 88Excellent quality (A+) · 88 of 100: how capable the model is at FP16, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. 62.3 min Very slow
Bernini-R A14B (video generation + editing)ByteDance · 14.3B 83Great quality (A) · 83 of 100: how capable the model is at FP8, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking. 35.2 min Slow 85Excellent quality (A+) · 85 of 100: how capable the model is at FP16, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 8.2 h Very slow
HY-OmniWeaving (8.3B, unified video)Tencent · 8.3B 74Great quality (A) · 74 of 100: how capable the model is at FP8, whatever it runs on. Match score 78/100 (quality, fit, speed and memory use) sets the ranking. 25.9 min Slow 76Great quality (A) · 76 of 100: how capable the model is at FP16, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. 6.1 h Very slow
SCAIL-2 14B (character animation)Zhipu AI · 16.4B 78Great quality (A) · 78 of 100: how capable the model is at FP8, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking. 58.7 min Very slowtight fit 80Great quality (A) · 80 of 100: how capable the model is at FP16, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 13.7 h Very slow
Wan 2.1 I2V 14B 480PAlibaba · 14.0B 72Great quality (A) · 72 of 100: how capable the model is at FP8, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking. 44.0 min Very slow 74Great quality (A) · 74 of 100: how capable the model is at FP16, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 10.3 h Very slow
FramePack I2V (HunyuanVideo 13B)lllyasviel · 12.9B 72Great quality (A) · 72 of 100: how capable the model is at FP8, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking. 19.6 min Slowtight fit 74Great quality (A) · 74 of 100: how capable the model is at FP16, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. 4.6 h Very slow
LTX-2.3Lightricks · 18.9B 79Great quality (A) · 79 of 100: how capable the model is at Q4 GGUF, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking. 27.4 min Slowencoder in RAM 86Excellent quality (A+) · 86 of 100: how capable the model is at FP16, whatever it runs on. Match score 68/100 (quality, fit, speed and memory use) sets the ranking. 4.6 h Very slow
Wan 2.2 S2V 14B (speech to video)Alibaba · 16.3B 74Great quality (A) · 74 of 100: how capable the model is at FP8, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking. 48.9 min Very slowtight fit 76Great quality (A) · 76 of 100: how capable the model is at FP16, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. 11.4 h Very slow
LTX-Video 2BLightricks · 1.9B 50Fair quality (C) · 50 of 100: how capable the model is at FP16, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking. 82 s Runs great 50Fair quality (C) · 50 of 100: how capable the model is at FP16, whatever it runs on. Match score 54/100 (quality, fit, speed and memory use) sets the ranking. 17.1 min Slow
LTX-2 (19B, video + audio)Lightricks · 18.9B 76Great quality (A) · 76 of 100: how capable the model is at Q4 GGUF, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. 27.4 min Slowencoder in RAM 82Great quality (A) · 82 of 100: how capable the model is at FP16, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 4.6 h Very slow
SkyReels V2 T2V 14B 540PSkywork · 14.0B 72Great quality (A) · 72 of 100: how capable the model is at FP8, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. 68.5 min Very slowtight fit 74Great quality (A) · 74 of 100: how capable the model is at FP16, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 16.0 h Very slow
Wan 2.2 I2V A14BAlibaba · 14.0B 79Great quality (A) · 79 of 100: how capable the model is at Q4 GGUF, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. 1.8 h Very slowencoder in RAM 86Excellent quality (A+) · 86 of 100: how capable the model is at FP16, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 18.3 h Very slow
Wan 2.2 T2V A14BAlibaba · 14.0B 79Great quality (A) · 79 of 100: how capable the model is at Q4 GGUF, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. 1.8 h Very slowencoder in RAM 86Excellent quality (A+) · 86 of 100: how capable the model is at FP16, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 18.3 h Very slow
CogVideoX 5BZhipu AI · 5.0B 55Good quality (B) · 55 of 100: how capable the model is at FP16, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. 9.8 min Runs well 55Good quality (B) · 55 of 100: how capable the model is at FP16, whatever it runs on. Match score 53/100 (quality, fit, speed and memory use) sets the ranking. 2.1 h Very slow
CogVideoX 5B I2VZhipu AI · 5.6B 55Good quality (B) · 55 of 100: how capable the model is at FP16, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. 9.8 min Runs well 55Good quality (B) · 55 of 100: how capable the model is at FP16, whatever it runs on. Match score 53/100 (quality, fit, speed and memory use) sets the ranking. 2.1 h Very slow
Cosmos-Predict2 2B Video2WorldNVIDIA · 2.0B 56Good quality (B) · 56 of 100: how capable the model is at FP8, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. 12.2 min Slow 58Good quality (B) · 58 of 100: how capable the model is at FP16, whatever it runs on. Match score 54/100 (quality, fit, speed and memory use) sets the ranking. 2.9 h Very slow
HunyuanVideo 1.5 (8.3B)Tencent · 8.3B 82Great quality (A) · 82 of 100: how capable the model is at FP8, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. 1.5 h Very slowtight fit · encoder in RAM 84Great quality (A) · 84 of 100: how capable the model is at FP16, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 20.5 h Very slow
Wan 2.2 TI2V 5BAlibaba · 5.0B 65Good quality (B) · 65 of 100: how capable the model is at Q4 GGUF, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. 33.9 min Slowtight fit 70Great quality (A) · 70 of 100: how capable the model is at FP16, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 5.9 h Very slow
Wan 2.1 VACE 1.3BAlibaba · 2.2B 55Good quality (B) · 55 of 100: how capable the model is at FP16, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. 14.1 min Slow 55Good quality (B) · 55 of 100: how capable the model is at FP16, whatever it runs on. Match score 52/100 (quality, fit, speed and memory use) sets the ranking. 3.0 h Very slow
Mochi 1 Preview (10B)Genmo · 10.0B 60Good quality (B) · 60 of 100: how capable the model is at FP8, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking. 53.8 min Very slow 62Good quality (B) · 62 of 100: how capable the model is at FP16, whatever it runs on. Match score 55/100 (quality, fit, speed and memory use) sets the ranking. 12.6 h Very slow
MiniMax H3 (33B, video + audio)MiniMax · 20.1B 89Excellent quality (A+) · 89 of 100: how capable the model is at Q8 GGUF, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 84.5 min Very slowCPU offload 90Excellent quality (A+) · 90 of 100: how capable the model is at FP16, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. 11.9 h Very slow
Wan 2.1 I2V 14B 720PAlibaba · 14.0B 72Great quality (A) · 72 of 100: how capable the model is at Q4 GGUF, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 2.3 h Very slowencoder in RAM 78Great quality (A) · 78 of 100: how capable the model is at FP16, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 22.8 h Very slow
Wan 2.1 T2V 14BAlibaba · 14.0B 72Great quality (A) · 72 of 100: how capable the model is at Q4 GGUF, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 2.3 h Very slowencoder in RAM 78Great quality (A) · 78 of 100: how capable the model is at FP16, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 22.8 h Very slow
SkyReels V2 Diffusion Forcing 1.3BSkywork · 1.4B 52Fair quality (C) · 52 of 100: how capable the model is at FP16, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 16.3 min Slow 52Fair quality (C) · 52 of 100: how capable the model is at FP16, whatever it runs on. Match score 51/100 (quality, fit, speed and memory use) sets the ranking. 3.4 h Very slow
Wan 2.2 Animate 14BAlibaba · 17.3B 72Great quality (A) · 72 of 100: how capable the model is at Q4 GGUF, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 2.2 h Very slowtight fit · encoder in RAM 78Great quality (A) · 78 of 100: how capable the model is at FP16, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 21.7 h Very slow
CogVideoX 2BZhipu AI · 1.7B 44Fair quality (C) · 44 of 100: how capable the model is at FP8, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 2.9 min Runs great 44Fair quality (C) · 44 of 100: how capable the model is at FP8, whatever it runs on. Match score 49/100 (quality, fit, speed and memory use) sets the ranking. 37.0 min Slow
Wan 2.1 VACE 14BAlibaba · 17.3B 72Great quality (A) · 72 of 100: how capable the model is at Q4 GGUF, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 2.5 h Very slowtight fit · encoder in RAM 78Great quality (A) · 78 of 100: how capable the model is at FP16, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 25.1 h Very slow
Kandinsky 6.0 Video Lite (3B, video + audio)Sber AI · 3.2B 68Good quality (B) · 68 of 100: how capable the model is at FP16, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 17.7 min Slowencoder in RAM 68Good quality (B) · 68 of 100: how capable the model is at FP16, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. 3.5 h Very slow
Kandinsky 6.0 Pro (29B, video + audio)Sber AI · 30.1B 88Excellent quality (A+) · 88 of 100: how capable the model is at FP16, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 3.5 h Very slowtight fit · CPU offload 88Excellent quality (A+) · 88 of 100: how capable the model is at FP16, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 33.6 h Very slow
Wan 2.1 T2V 1.3BAlibaba · 1.3B 53Fair quality (C) · 53 of 100: how capable the model is at FP8, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 11.7 min Runs well 55Good quality (B) · 55 of 100: how capable the model is at FP16, whatever it runs on. Match score 52/100 (quality, fit, speed and memory use) sets the ranking. 2.7 h Very slow
HunyuanVideo 13BTencent · 13.0B 66Good quality (B) · 66 of 100: how capable the model is at Q4 GGUF, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 2.7 h Very slowtight fit · encoder in RAM 72Great quality (A) · 72 of 100: how capable the model is at FP16, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 27.4 h Very slow
HunyuanVideo I2V 13BTencent · 13.0B 66Good quality (B) · 66 of 100: how capable the model is at Q4 GGUF, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 2.7 h Very slowtight fit · encoder in RAM 72Great quality (A) · 72 of 100: how capable the model is at FP16, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. 27.4 h Very slow
SkyReels V2 T2V 14B 720PSkywork · 14.0B 76Great quality (A) · 76 of 100: how capable the model is at FP8, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 2.5 h Very slowCPU offload 78Great quality (A) · 78 of 100: how capable the model is at FP16, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 27.4 h Very slow
Step-Video T2V (30B)StepFun · 30.0B 74Great quality (A) · 74 of 100: how capable the model is at FP8, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. 3.3 h Very slowCPU offload 76Great quality (A) · 76 of 100: how capable the model is at FP16, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 35.4 h Very slow
Cosmos-Predict2 14B Video2WorldNVIDIA · 14.0B 68Good quality (B) · 68 of 100: how capable the model is at FP8, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 1.7 h Very slowCPU offload 70Great quality (A) · 70 of 100: how capable the model is at FP16, whatever it runs on. Match score 60/100 (quality, fit, speed and memory use) sets the ranking. 18.3 h Very slow
Allegro (2.8B)Rhymes AI · 2.8B 52Fair quality (C) · 52 of 100: how capable the model is at FP16, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 68.5 min Very slowtight fit · encoder in RAM 52Fair quality (C) · 52 of 100: how capable the model is at FP16, whatever it runs on. Match score 51/100 (quality, fit, speed and memory use) sets the ranking. 13.7 h Very slow
Open-Sora 2.0 (11B)HPC-AI Tech · 11.0B 70Great quality (A) · 70 of 100: how capable the model is at FP16, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 3.5 h Very slowCPU offload 70Great quality (A) · 70 of 100: how capable the model is at FP16, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 34.2 h Very slow
AnimateDiff v3 (SD 1.5)Shanghai AI Lab · 1.3B 35Basic quality (D) · 35 of 100: how capable the model is at FP16, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. 65 s Runs great 35Basic quality (D) · 35 of 100: how capable the model is at FP16, whatever it runs on. Match score 48/100 (quality, fit, speed and memory use) sets the ranking. 13.7 min Slow
Cosmos 1.0 Diffusion 7B Text2WorldNVIDIA · 7.0B 60Good quality (B) · 60 of 100: how capable the model is at FP16, whatever it runs on. Match score 56/100 (quality, fit, speed and memory use) sets the ranking. 38.9 min Slowtight fit · CPU offload 60Good quality (B) · 60 of 100: how capable the model is at FP16, whatever it runs on. Match score 55/100 (quality, fit, speed and memory use) sets the ranking. 6.3 h Very slow
Cosmos3-Super Image2Video (64B)NVIDIA · 64.0B 88Excellent quality (A+) · 88 of 100: how capable the model is at FP16, whatever it runs on. Match score 11/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 88Excellent quality (A+) · 88 of 100: how capable the model is at FP16, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. 25.7 h Very slow
Wan-Dancer 14B (music to dance)Alibaba · 17.3B 74Great quality (A) · 74 of 100: how capable the model is at FP8, whatever it runs on. Match score 10/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 76Great quality (A) · 76 of 100: how capable the model is at FP16, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 50.2 h Very slow
CogVideoX 1.5 5BZhipu AI · 5.6B 58Good quality (B) · 58 of 100: how capable the model is at FP8, whatever it runs on. Match score 7/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 60Good quality (B) · 60 of 100: how capable the model is at FP16, whatever it runs on. Match score 55/100 (quality, fit, speed and memory use) sets the ranking. 10.3 h Very slow

Decision models

68 of 70 decision models run on the Intel Arc Pro B70. 32.0 GB usable for models.

70 of 70 decision models run on the AMD EPYC 7002 / 7003 series NAS (512 GB). 471 GB usable for models.

Model QualityArc Pro B70 SpeedArc Pro B70 RunsArc Pro B70 QualityEPYC 7002 / 7003 series NAS SpeedEPYC 7002 / 7003 series NAS RunsEPYC 7002 / 7003 series NAS
Qwen3 Reranker 4BAlibaba · 4.0B 90Excellent quality (A+) · 90 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 94/100 (quality, fit, speed and memory use) sets the ranking. 118 ms Runs great 90Excellent quality (A+) · 90 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 82/100 (quality, fit, speed and memory use) sets the ranking. 5.2 s Very slow
Jina Reranker v3.5Jina AI · 600M 86Excellent quality (A+) · 86 of 100: how capable the model is at F16, whatever it runs on. Match score 93/100 (quality, fit, speed and memory use) sets the ranking. 23 ms Runs great 86Excellent quality (A+) · 86 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 84/100 (quality, fit, speed and memory use) sets the ranking. 780 ms Slow
zerank-2ZeroEntropy · 4.0B 88Excellent quality (A+) · 88 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 93/100 (quality, fit, speed and memory use) sets the ranking. 118 ms Runs great 88Excellent quality (A+) · 88 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking. 5.2 s Very slow
Shieldstral 1.0 3BMistral AI · 3.9B 86Excellent quality (A+) · 86 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 92/100 (quality, fit, speed and memory use) sets the ranking. 113 ms Runs great 86Excellent quality (A+) · 86 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. 5.0 s Very slow
Intern-Decision 4BShanghai AI Lab · 4.5B 84Great quality (A) · 84 of 100: how capable the model is at F16, whatever it runs on. Match score 92/100 (quality, fit, speed and memory use) sets the ranking. 94 ms Runs great 84Great quality (A) · 84 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. 3.0 s Very slow
Jina Reranker v3Jina AI · 600M 84Great quality (A) · 84 of 100: how capable the model is at F16, whatever it runs on. Match score 92/100 (quality, fit, speed and memory use) sets the ranking. 23 ms Runs great 84Great quality (A) · 84 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 83/100 (quality, fit, speed and memory use) sets the ranking. 780 ms Slow
d1 3BLiquid AI · 3.1B 82Great quality (A) · 82 of 100: how capable the model is at F16, whatever it runs on. Match score 91/100 (quality, fit, speed and memory use) sets the ranking. 66 ms Runs great 82Great quality (A) · 82 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. 2.0 s Very slow
mxbai-rerank-large v2Mixedbread · 1.5B 82Great quality (A) · 82 of 100: how capable the model is at F16, whatever it runs on. Match score 91/100 (quality, fit, speed and memory use) sets the ranking. 54 ms Runs great 82Great quality (A) · 82 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. 2.0 s Very slow
FRIDA-DecisionsSber AI · 820M 80Great quality (A) · 80 of 100: how capable the model is at F16, whatever it runs on. Match score 91/100 (quality, fit, speed and memory use) sets the ranking. 17 ms Runs great 80Great quality (A) · 80 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 82/100 (quality, fit, speed and memory use) sets the ranking. 574 ms Slow
GLiNER2.5 DecideFastino · 490M 80Great quality (A) · 80 of 100: how capable the model is at F16, whatever it runs on. Match score 91/100 (quality, fit, speed and memory use) sets the ranking. 12 ms Runs great 80Great quality (A) · 80 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 84/100 (quality, fit, speed and memory use) sets the ranking. 345 ms Runs well
LemonDecider 4BAnt Group (inclusionAI) · 4.5B 80Great quality (A) · 80 of 100: how capable the model is at F16, whatever it runs on. Match score 91/100 (quality, fit, speed and memory use) sets the ranking. 75 ms Runs great 80Great quality (A) · 80 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 78/100 (quality, fit, speed and memory use) sets the ranking. 3.2 s Very slow
Qwen3 Reranker 0.6BAlibaba · 600M 80Great quality (A) · 80 of 100: how capable the model is at F16, whatever it runs on. Match score 91/100 (quality, fit, speed and memory use) sets the ranking. 23 ms Runs great 80Great quality (A) · 80 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking. 780 ms Slow
zerank-1 smallZeroEntropy · 1.7B 80Great quality (A) · 80 of 100: how capable the model is at F16, whatever it runs on. Match score 91/100 (quality, fit, speed and memory use) sets the ranking. 60 ms Runs great 80Great quality (A) · 80 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. 2.2 s Very slow
Qwen3 Reranker 8BAlibaba · 8.2B 90Excellent quality (A+) · 90 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 90/100 (quality, fit, speed and memory use) sets the ranking. 236 ms Runs well 90Excellent quality (A+) · 90 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking. 10.6 s Very slow
BGE Reranker v2 GemmaBAAI · 2.5B 78Great quality (A) · 78 of 100: how capable the model is at F16, whatever it runs on. Match score 90/100 (quality, fit, speed and memory use) sets the ranking. 85 ms Runs great 78Great quality (A) · 78 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking. 3.3 s Very slow
Llama Prompt Guard 2 86MMeta · 280M 76Great quality (A) · 76 of 100: how capable the model is at F16, whatever it runs on. Match score 89/100 (quality, fit, speed and memory use) sets the ranking. 13 ms Runs great 76Great quality (A) · 76 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking. 394 ms Runs well
mxbai-rerank-base v2Mixedbread · 490M 76Great quality (A) · 76 of 100: how capable the model is at F16, whatever it runs on. Match score 89/100 (quality, fit, speed and memory use) sets the ranking. 20 ms Runs great 76Great quality (A) · 76 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. 638 ms Slow
GLiClass ModernBERT Large v3Knowledgator · 400M 74Great quality (A) · 74 of 100: how capable the model is at F16, whatever it runs on. Match score 88/100 (quality, fit, speed and memory use) sets the ranking. 10 ms Runs great 74Great quality (A) · 74 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 82/100 (quality, fit, speed and memory use) sets the ranking. 282 ms Runs well
OpenAI Privacy FilterOpenAI · 1.4B (50M active) 78Great quality (A) · 78 of 100: how capable the model is at F16, whatever it runs on. Match score 88/100 (quality, fit, speed and memory use) sets the ranking. 4.8 ms Runs great 78Great quality (A) · 78 of 100: how capable the model is at F16, whatever it runs on. Match score 88/100 (quality, fit, speed and memory use) sets the ranking. 44 ms Runs great
Skywork Reward V2 Llama 3.1 8BSkywork · 7.5B 90Excellent quality (A+) · 90 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 87/100 (quality, fit, speed and memory use) sets the ranking. 410 ms Slow 90Excellent quality (A+) · 90 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking. 20.9 s Very slow
BGE Reranker v2 M3BAAI · 570M 72Great quality (A) · 72 of 100: how capable the model is at F16, whatever it runs on. Match score 87/100 (quality, fit, speed and memory use) sets the ranking. 22 ms Runs great 72Great quality (A) · 72 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking. 797 ms Slow
DeBERTa v3 Large Zero-shot v2.0Moritz Laurer · 440M 72Great quality (A) · 72 of 100: how capable the model is at F16, whatever it runs on. Match score 87/100 (quality, fit, speed and memory use) sets the ranking. 11 ms Runs great 72Great quality (A) · 72 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. 310 ms Runs well
DuoGuard 0.5BDuoGuard · 490M 72Great quality (A) · 72 of 100: how capable the model is at F16, whatever it runs on. Match score 87/100 (quality, fit, speed and memory use) sets the ranking. 19 ms Runs great 72Great quality (A) · 72 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 78/100 (quality, fit, speed and memory use) sets the ranking. 686 ms Slow
Skywork Reward V2 Qwen3 0.6BSkywork · 600M 72Great quality (A) · 72 of 100: how capable the model is at F16, whatever it runs on. Match score 87/100 (quality, fit, speed and memory use) sets the ranking. 41 ms Runs great 72Great quality (A) · 72 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking. 1.7 s Very slow
GTE Reranker ModernBERT BaseAlibaba · 150M 70Great quality (A) · 70 of 100: how capable the model is at F16, whatever it runs on. Match score 86/100 (quality, fit, speed and memory use) sets the ranking. 8.7 ms Runs great 70Great quality (A) · 70 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking. 213 ms Runs well
Jina Reranker v2 Base MultilingualJina AI · 280M 70Great quality (A) · 70 of 100: how capable the model is at F16, whatever it runs on. Match score 86/100 (quality, fit, speed and memory use) sets the ranking. 13 ms Runs great 70Great quality (A) · 70 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. 394 ms Runs well
ModernBERT Large Zero-shot v2.0Moritz Laurer · 400M 70Great quality (A) · 70 of 100: how capable the model is at F16, whatever it runs on. Match score 86/100 (quality, fit, speed and memory use) sets the ranking. 10 ms Runs great 70Great quality (A) · 70 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. 282 ms Runs well
GLiGuard 300MFastino · 210M 68Good quality (B) · 68 of 100: how capable the model is at F16, whatever it runs on. Match score 85/100 (quality, fit, speed and memory use) sets the ranking. 11 ms Runs great 68Good quality (B) · 68 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. 296 ms Runs well
Granite Embedding Reranker English R2IBM · 150M 68Good quality (B) · 68 of 100: how capable the model is at F16, whatever it runs on. Match score 85/100 (quality, fit, speed and memory use) sets the ranking. 8.7 ms Runs great 68Good quality (B) · 68 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. 213 ms Runs well
LayaConvai Innovations · 420M 68Good quality (B) · 68 of 100: how capable the model is at F16, whatever it runs on. Match score 85/100 (quality, fit, speed and memory use) sets the ranking. 11 ms Runs great 68Good quality (B) · 68 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. 296 ms Runs well
Llama Prompt Guard 2 22MMeta · 71M 68Good quality (B) · 68 of 100: how capable the model is at F16, whatever it runs on. Match score 85/100 (quality, fit, speed and memory use) sets the ranking. 6.2 ms Runs great 68Good quality (B) · 68 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 85/100 (quality, fit, speed and memory use) sets the ranking. 103 ms Runs great
Qwen3Guard Gen 4BAlibaba · 4.4B 83Great quality (A) · 83 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 85/100 (quality, fit, speed and memory use) sets the ranking. 181 ms Runs well 84Great quality (A) · 84 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking. 6.0 s Very slow
Qwen3Guard Gen 0.6BAlibaba · 750M 74Great quality (A) · 74 of 100: how capable the model is at F16, whatever it runs on. Match score 85/100 (quality, fit, speed and memory use) sets the ranking. 56 ms Runs great 74Great quality (A) · 74 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. 1.0 s Slow
Skywork Reward V2 Qwen3 8BSkywork · 7.6B 84Great quality (A) · 84 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 85/100 (quality, fit, speed and memory use) sets the ranking. 414 ms Slow 84Great quality (A) · 84 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 78/100 (quality, fit, speed and memory use) sets the ranking. 21.1 s Very slow
GLiClass Base v3Knowledgator · 190M 66Good quality (B) · 66 of 100: how capable the model is at F16, whatever it runs on. Match score 84/100 (quality, fit, speed and memory use) sets the ranking. 7.0 ms Runs great 66Good quality (B) · 66 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 82/100 (quality, fit, speed and memory use) sets the ranking. 136 ms Runs well
ProtectAI DeBERTa v3 Prompt Injection v2Protect AI · 180M 66Good quality (B) · 66 of 100: how capable the model is at F16, whatever it runs on. Match score 84/100 (quality, fit, speed and memory use) sets the ranking. 9.6 ms Runs great 66Good quality (B) · 66 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. 254 ms Runs well
DeBERTa v3 Base Zero-shot v2.0Moritz Laurer · 180M 64Good quality (B) · 64 of 100: how capable the model is at F16, whatever it runs on. Match score 83/100 (quality, fit, speed and memory use) sets the ranking. 6.8 ms Runs great 64Good quality (B) · 64 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking. 129 ms Runs great
GLiNER2 Base v1Fastino · 210M 68Good quality (B) · 68 of 100: how capable the model is at F16, whatever it runs on. Match score 83/100 (quality, fit, speed and memory use) sets the ranking. 7.3 ms Runs great 68Good quality (B) · 68 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. 150 ms Runs well
Qwen3Guard Gen 8BAlibaba · 8.2B 85Excellent quality (A+) · 85 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 83/100 (quality, fit, speed and memory use) sets the ranking. 332 ms Runs well 86Excellent quality (A+) · 86 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking. 11.2 s Very slow
HHEM 2.1 OpenVectara · 110M 62Good quality (B) · 62 of 100: how capable the model is at F16, whatever it runs on. Match score 82/100 (quality, fit, speed and memory use) sets the ranking. 11 ms Runs great 62Good quality (B) · 62 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking. 310 ms Runs well
Llama 3.1 Nemotron Safety Guard 8B v3NVIDIA · 8.0B 83Great quality (A) · 83 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 82/100 (quality, fit, speed and memory use) sets the ranking. 326 ms Runs well 84Great quality (A) · 84 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking. 11.0 s Very slow
LettuceDetect LargeKR Labs · 400M 66Good quality (B) · 66 of 100: how capable the model is at F16, whatever it runs on. Match score 82/100 (quality, fit, speed and memory use) sets the ranking. 29 ms Runs great 66Good quality (B) · 66 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. 1.1 s Slow
Granite Guardian 4.1 8BIBM · 8.4B 83Great quality (A) · 83 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 82/100 (quality, fit, speed and memory use) sets the ranking. 340 ms Runs well 84Great quality (A) · 84 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking. 11.4 s Very slow
Laya MultilingualConvai Innovations · 320M 60Good quality (B) · 60 of 100: how capable the model is at F16, whatever it runs on. Match score 82/100 (quality, fit, speed and memory use) sets the ranking. 9.0 ms Runs great 60Good quality (B) · 60 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking. 227 ms Runs well
WildGuard 7BAi2 · 7.3B 79Great quality (A) · 79 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking. 294 ms Runs well 80Great quality (A) · 80 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. 9.9 s Very slow
NuExtract 2.0 2BNuMind · 2.2B 66Good quality (B) · 66 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking. 125 ms Runs great 70Great quality (A) · 70 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. 2.2 s Very slow
ShieldGemma 2BGoogle · 2.6B 58Good quality (B) · 58 of 100: how capable the model is at F16, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking. 88 ms Runs great 58Good quality (B) · 58 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 68/100 (quality, fit, speed and memory use) sets the ranking. 3.4 s Very slow
GLiNER Multi v2.1Urchade Zaratiana · 290M 62Good quality (B) · 62 of 100: how capable the model is at F16, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. 8.5 ms Runs great 62Good quality (B) · 62 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 76/100 (quality, fit, speed and memory use) sets the ranking. 206 ms Runs well
Granite Guardian 3.3 8BIBM · 8.2B 79Great quality (A) · 79 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. 331 ms Runs well 80Great quality (A) · 80 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. 11.2 s Very slow
NuExtract 2.0 4BNuMind · 3.8B 72Great quality (A) · 72 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. 209 ms Runs well 76Great quality (A) · 76 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. 3.7 s Very slow
BART Large MNLIMeta · 410M 56Good quality (B) · 56 of 100: how capable the model is at F16, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. 10 ms Runs great 56Good quality (B) · 56 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. 289 ms Runs well
ms-marco MiniLM L6 v2Sentence Transformers · 23M 52Fair quality (C) · 52 of 100: how capable the model is at F16, whatever it runs on. Match score 78/100 (quality, fit, speed and memory use) sets the ranking. 4.7 ms Runs great 52Fair quality (C) · 52 of 100: how capable the model is at F16, whatever it runs on. Match score 78/100 (quality, fit, speed and memory use) sets the ranking. 41 ms Runs great
ShieldGemma 9BGoogle · 9.2B 64Good quality (B) · 64 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 78/100 (quality, fit, speed and memory use) sets the ranking. 266 ms Runs well 64Good quality (B) · 64 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking. 12.0 s Very slow
Llama Guard 3 1BMeta · 1.5B 56Good quality (B) · 56 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking. 72 ms Runs great 56Good quality (B) · 56 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 2.1 s Very slow
Llama Guard 3 8BMeta · 8.0B 67Good quality (B) · 67 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 75/100 (quality, fit, speed and memory use) sets the ranking. 326 ms Runs well 68Good quality (B) · 68 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 11.0 s Very slow
ShieldGemma 27BGoogle · 27B 65Good quality (B) · 65 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 74/100 (quality, fit, speed and memory use) sets the ranking. 750 ms Slow 66Good quality (B) · 66 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking. 35.2 s Very slow
Llama Guard 4 12BMeta · 12B 65Good quality (B) · 65 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. 485 ms Slow 66Good quality (B) · 66 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 16.4 s Very slow
gpt-oss-safeguard 20BOpenAI · 22B (3.6B active) 76Great quality (A) · 76 of 100: how capable the model is at MXFP4, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. 1.6 s Very slow 76Great quality (A) · 76 of 100: how capable the model is at MXFP4, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking. 9.0 s Very slow
CompassJudger-2 32BOpenCompass · 33B 83Great quality (A) · 83 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. 12.1 s Very slow 86Excellent quality (A+) · 86 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. 153.0 s Very slow
DynaGuard 8BUniversity of Maryland · 8.2B 79Great quality (A) · 79 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. 4.7 s Very slow 79Great quality (A) · 79 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. 28.9 s Very slow
Nemotron Content Safety Reasoning 4BNVIDIA · 4.3B 74Great quality (A) · 74 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. 1.3 s Slow 78Great quality (A) · 78 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. 15.2 s Very slow
Selene 1 Mini 8BAtla · 8.0B 80Great quality (A) · 80 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. 4.6 s Very slow 80Great quality (A) · 80 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking. 37.5 s Very slow
CompassJudger-2 7BOpenCompass · 7.6B 78Great quality (A) · 78 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. 4.3 s Very slow 78Great quality (A) · 78 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. 35.6 s Very slow
DynaGuard 1.7BUniversity of Maryland · 2.0B 69Good quality (B) · 69 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking. 876 ms Slow 70Great quality (A) · 70 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 68/100 (quality, fit, speed and memory use) sets the ranking. 7.2 s Very slow
Prometheus 2 7BPrometheus Eval · 7.2B 67Good quality (B) · 67 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 3.1 s Very slow 68Good quality (B) · 68 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 33.8 s Very slow
Prometheus 2 8x7BPrometheus Eval · 47B (13B active) 70Great quality (A) · 70 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 4.5 s Very slowtight fit 74Great quality (A) · 74 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. 60.8 s Very slow
GLIDERPatronus AI · 3.8B 62Good quality (B) · 62 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. 1.2 s Slow 66Good quality (B) · 66 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. 17.8 s Very slow
Flow Judge v0.1Flow AI · 3.8B 60Good quality (B) · 60 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. 1.2 s Slow 64Good quality (B) · 64 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. 17.8 s Very slow
GLiDE 27BFastino · 26B 86Excellent quality (A+) · 86 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 11/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 86Excellent quality (A+) · 86 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. 18.2 s Very slow
gpt-oss-safeguard 120BOpenAI · 117B (5.1B active) 80Great quality (A) · 80 of 100: how capable the model is at MXFP4, whatever it runs on. Match score 10/100 (quality, fit, speed and memory use) sets the ranking. — Won't run 80Great quality (A) · 80 of 100: how capable the model is at MXFP4, whatever it runs on. Match score 72/100 (quality, fit, speed and memory use) sets the ranking. 12.5 s Very slow

Estimates come from published hardware specs and our open formulas, not from running the model on your machine. Real speed varies with runtime, drivers and settings. Numbers marked “Measured” come from a linked benchmark.

Was this page helpful?