LLMs · Mistral AI

Mixtral 8x22B Instruct

Name: Mixtral 8x22B Instruct
Author: Mistral AI

Parameters: 141B (39.0B active)
Architecture: Mixture of experts
Max context: 64K tokens
License: apache-2.0
Released: 2024-04

How to run it

Quickest · Ollama

One command downloads a ready-to-run version and starts a chat.

ollama run mixtral:8x22b

Get Ollama at ollama.com/download (opens in a new tab) · model page (opens in a new tab)

Quantized files · GGUF

Works in LM Studio, llama.cpp, Jan and Ollama. No sign-in needed.

huggingface.co/MaziyarPanahi/Mixtral-8x22B-Instruct-v0.1-GGUF (opens in a new tab)

ollama run hf.co/MaziyarPanahi/Mixtral-8x22B-Instruct-v0.1-GGUF

Original weights · Hugging Face

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

huggingface.co/mistralai/Mixtral-8x22B-Instruct-v0.1 (opens in a new tab)

No sign-in needed.

Official repo · GitHub

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

github.com/mistralai/mistral-inference (opens in a new tab)

git clone https://github.com/mistralai/mistral-inference

Mistral AI’s official inference code.

Links checked 7 days ago against Hugging Face, Ollama and GitHub.

Memory needed

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

Quantization	Quality	Weights	Total at 8K context	Total at 32K context	Download
`Q2_K`	Low	56.3 GB	58.7 GB	64.4 GB	GGUF · part 1 of 3 (opens in a new tab)
`Q3_K_M`	Low	68.9 GB	71.4 GB	77.0 GB	GGUF · part 1 of 2 (opens in a new tab)
`Q4_K_M`	Good	85.8 GB	88.3 GB	93.9 GB	GGUF · part 1 of 2 (opens in a new tab)
`Q5_K_M`	High	99.8 GB	102 GB	108 GB	GGUF · part 1 of 4 (opens in a new tab)
`Q6_K`	High	115 GB	118 GB	123 GB
`Q8_0`	Very high	149 GB	152 GB	157 GB	GGUF · part 1 of 4 (opens in a new tab)
`F16`	Full	281 GB	284 GB	289 GB	GGUF · part 1 of 6 (opens in a new tab) Original weights (opens in a new tab)

Best hardware for Mixtral 8x22B Instruct

The least expensive GPU in our catalog that runs it without offloading is the NVIDIA RTX PRO 6000 Blackwell (launch price $8,565) at Q3_K_M.

Grade · points	Hardware	Quant	Memory	Speed	Fit
75 ptsGreat (A) · 75 of 100 points: Model quality 20.8/45, Memory fit 25.0/25, Speed 14.3/15, Memory use 15.0/15	NVIDIA RTX PRO 6000 Blackwell	`Q3_K_M`	76.2 GB / 96.0 GB	36.5 tok/s	Runs well
71 ptsGood (B) · 71 of 100 points: Model quality 23.5/45, Memory fit 25.0/25, Speed 7.8/15, Memory use 15.0/15	Apple M2 Ultra (192 GB)	`Q4_K_M`	99.8 GB / 144 GB	10.8 tok/s	Runs well
68 ptsGood (B) · 68 of 100 points: Model quality 23.5/45, Memory fit 24.0/25, Speed 6.5/15, Memory use 15.0/15	Apple M4 Max (128 GB)	`Q4_K_M`	95.9 GB / 96.0 GB	7.4 tok/s	Tight fit
68 ptsGood (B) · 68 of 100 points: Model quality 23.5/45, Memory fit 25.0/25, Speed 7.9/15, Memory use 12.5/15	Apple M3 Ultra (256 GB)	`Q4_K_M`	104 GB / 192 GB	11.1 tok/s	Runs well
68 ptsGood (B) · 68 of 100 points: Model quality 23.5/45, Memory fit 25.0/25, Speed 4.9/15, Memory use 15.0/15	NVIDIA GB10 (128 GB)	`Q4_K_M`	97.5 GB / 115 GB	4.3 tok/s	Runs well
66 ptsGood (B) · 66 of 100 points: Model quality 23.5/45, Memory fit 24.0/25, Speed 4.4/15, Memory use 15.0/15	AMD Ryzen AI Max+ 395 (128 GB)	`Q4_K_M`	95.9 GB / 96.0 GB	3.5 tok/s	Tight fit
63 ptsGood (B) · 63 of 100 points: Model quality 23.5/45, Memory fit 25.0/25, Speed 7.9/15, Memory use 7.2/15	Apple M3 Ultra (512 GB)	`Q4_K_M`	119 GB / 384 GB	11.1 tok/s	Runs well

Check it on your own hardware

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.