LLMs ·

Step 5 Preview (600B-A27B)

Parameters
604B (27.0B active)
Architecture
Mixture of experts
Max context
1,024K tokens
License
stepfun-model-license-1.0
Released
2026-09

How to run it

Original weights · Hugging Face

Full-precision files from StepFun, for fine-tuning, vLLM, Transformers.

huggingface.co/stepfun-ai/Step-5-Preview-BF16 (opens in a new tab)

No sign-in needed.

Official repo · GitHub

Reference code, inference scripts and docs from StepFun. The weights themselves download from Hugging Face.

github.com/stepfun-ai/Step-5-Preview (opens in a new tab)

git clone https://github.com/stepfun-ai/Step-5-Preview

Links checked 5 days ago against Hugging Face, Ollama and GitHub.

Apps that run Step 5 Preview (600B-A27B)

Our pick for each kind of computer, assuming the model fits in its memory. Pick your own GPU or device in the Model Finder for an exact answer.

Mac (Apple silicon)

Docker Model Runner

Handy if you already work in Docker.

Also good:

Windows PC, NVIDIA GPU

Docker Model Runner

Handy if you already work in Docker.

Also good:

Windows PC, AMD Radeon

Docker Model Runner

Handy if you already work in Docker.

Also good:

Linux, NVIDIA GPU

Docker Model Runner

Handy if you already work in Docker.

Also good:

Linux server, 80 GB+ GPUs

Docker Model Runner

Handy if you already work in Docker.

Also good:

App Type Opens it as Systems Engine Start
vLLM Server Original weights WSLWindowsLinux vLLM 
vllm serve stepfun-ai/Step-5-Preview-BF16
SGLang Server Original weights WSLWindowsLinux SGLang 
python3 -m sglang.launch_server --model-path stepfun-ai/Step-5-Preview-BF16
Docker Model Runner CLI + API Original weights macOSWindowsLinux llama.cpp, vLLM 
docker model run hf.co/stepfun-ai/Step-5-Preview-BF16
Text Generation WebUI Desktop app Original weights macOSWindowsLinux llama.cpp, Transformers, ExLlamaV3 Get Text Generation WebUI

Which apps can open a model depends on the files published for it: GGUF for llama.cpp-based apps, MLX for Apple’s engine, the original weights for vLLM and SGLang. Compare every app.

Memory needed

Weights plus KV cache and runtime overhead. Leave a little extra free for the operating system and other apps. Try another context length.

KV cache: sliding-window hybrid: about 1 in 4 layers is global (window 512); the KV estimate counts the global-attention layers plus a small allowance for the sliding-window layers.

Quantization Quality Weights Total at 8K context Total at 32K context Download
Q2_K Low 242 GB 243 GB 245 GB All files (opens in a new tab)
Q3_K_M Low 296 GB 297 GB 299 GB All files (opens in a new tab)
Q4_K_M Good 369 GB 370 GB 372 GB All files (opens in a new tab)
Q5_K_M High 429 GB 430 GB 432 GB All files (opens in a new tab)
Q6_K High 496 GB 497 GB 498 GB All files (opens in a new tab)
Q8_0 Very high 641 GB 642 GB 644 GB All files (opens in a new tab)
F16 Full 1,209 GB 1,210 GB 1,212 GB Original weights (opens in a new tab)

Best hardware for Step 5 Preview (600B-A27B)

Quality Hardware Quant Memory Speed Runs
81Great quality (A) · 81 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 90/100 (quality, fit, speed and memory use) sets the ranking. Apple Mac Studio (2025) (M4 Max or M3 Ultra)Apple M3 Ultra (512 GB) Q3_K_M 328 GB / 384 GB 36.5 tok/s Runs welltight fit
81Great quality (A) · 81 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 90/100 (quality, fit, speed and memory use) sets the ranking. Apple Mac Studio (M5 Max or M5 Ultra)Apple M5 Ultra (512 GB) Q3_K_M 328 GB / 384 GB 50.7 tok/s Runs welltight fit
67Good quality (B) · 67 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 84/100 (quality, fit, speed and memory use) sets the ranking. AMD Instinct MI325XAMD Instinct MI325X Q2_K 256 GB / 256 GB 175 tok/s Runs welltight fit
67Good quality (B) · 67 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 84/100 (quality, fit, speed and memory use) sets the ranking. AMD Instinct MI350XAMD Instinct MI350X Q2_K 257 GB / 288 GB 197 tok/s Runs welltight fit
67Good quality (B) · 67 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. Dell Pro Max with GB300NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip Q2_K 256 GB / 252 GB 127 tok/s Runs welloffloads to RAM
94Excellent quality (A+) · 94 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. QNAP QAI-h1290FXAMD EPYC 7002 / 7003 series NAS (512 GB) Q5_K_M 468 GB / 471 GB 4.2 tok/s Slowtight fit
81Great quality (A) · 81 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. QNAP TDS-h6889AXU-RPIntel Xeon Silver (4100 to 4300 series) NAS (512 GB) Q3_K_M 335 GB / 471 GB 3.3 tok/s Slow
81Great quality (A) · 81 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. QNAP TS-2888XIntel Xeon W-2100 series NAS (512 GB) Q3_K_M 335 GB / 471 GB 3.1 tok/s Slow
67Good quality (B) · 67 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 73/100 (quality, fit, speed and memory use) sets the ranking. TerraMaster U8-725Intel Xeon D-2700 series NAS (512 GB) Q2_K 281 GB / 471 GB 3.1 tok/s Slow

Check it on your own hardware

Run it on a cluster

Devices that can’t run Step 5 Preview (600B-A27B) alone can run it together. These example clusters do:

Try it with your own devices

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?