Model

Mixtral 8x22B

OPEN

Mistral AI · Mixtral family · released Apr 17, 2024

Larger Apache-2.0 sparse MoE (141B total / 39B active) with a 64K context window and native function calling; strong multilingual, math and code.

ReasoningCodingVisionFunction callingTool useAgentic

678.2

Elo · rank #270

Parameters

141B

Active params

39B (MoE)

Context

64K tokens

Architecture

Sparse Mixture-of-Experts, 8 experts, top-2 routing

License

Apache 2.0

Languages

API price (in/out)

No hosted API

Modalities

text

Benchmark results

Bar shows position within the tracked field; marker = field best

BIG-Bench HardReasoning61.3%#77

best: ERNIE 4.5 300B-A47B · 94.3%

GPQA DiamondReasoning37.3%#232

best: GPT-5.6 · 94.6%

GSM8KMath90.8%#38

best: Llama 3.1 405B · 96.8%

IFEvalReasoning71.8%#111

best: Gemma 4 26B A4B · 98.5%

LogicKorHuman preference6.79#20

best: GPT-4o · 9.33

MATH-500Math44.6%#142

best: GPT-5 · 99.4%

MMLU-ProKnowledge44.8%#144

best: Claude Fable 5 · 91.5%

MMLUKnowledge77.3%#113

best: OpenAI o3 · 92.9%

Run it locally

VRAM @ Q4

80 GB

VRAM @ FP16

263 GB

Fits on (Q4)

M3 Max 128GBM3 Ultra 512GBH200 141GBB200 192GB

Does not fit a single RTX 4090 (24GB) even at Q4 (~80GB needed); requires multi-GPU or large unified-memory hardware

Quantizations

GGUF Q4 · AWQ · MLX

Mixtral family

Elo progression across releases

Mixtral 8x7BDec 2023438.9 Mixtral 8x22BApr 2024678.2

Resources

HFmistralai/Mixtral-8x22B-v0.1 GHmistralai/mistral-inference DOCSdocs.mistral.ai/models/mixtral-8x22b-0-1-0-3 APIAvailable via Mistral AI API

API price weights · each benchmark row carries its own source badge (see methodology)