Model
Model explorer

Mixtral 8x7B

OPEN
Mistral AI · Mixtral family · released Dec 11, 2023

First strong open sparse MoE (8 experts, top-2 routing); matched/beat GPT-3.5 at ~13B active params under Apache 2.0.

ReasoningCodingVisionFunction callingTool useAgentic
438.9
Elo · rank #322
Parameters
46.7B
Active params
12.9B (MoE)
Context
32K tokens
Architecture
Sparse Mixture-of-Experts, 8 experts, top-2 routing
License
Apache 2.0
Languages
5+
API price (in/out)
No hosted API
Modalities
text
Benchmark results
Bar shows position within the tracked field; marker = field best
ARC-ChallengeReasoning59.7%#93
best: Llama 3.1 405B · 96.9%
ARC-EasyReasoning83.1%#17
best: Phi-3-medium (14B) · 97.7%
GSM8KMath74.4%#92
best: Llama 3.1 405B · 96.8%
HellaSwagReasoning84.4%#48
best: Claude 3 Opus · 95.4%
HumanEvalCoding40.2%#138
best: Claude Opus 4.5 · 99.4%
LogicKorHuman preference5.98#26
best: GPT-4o · 9.33
MATH-500Math28.4%#168
best: GPT-5 · 99.4%
MBPPCoding60.7%#58
best: Llama-3.3-Nemotron-Super-49B v1 (Reasoning On) · 91.3%
MMLU (EU-21 languages)Knowledge61.5%#2
best: Llama 3.1 70B · 77.1%
MMLUKnowledge70.6%#141
best: OpenAI o3 · 92.9%
MT-BenchHuman preference8.3#38
best: Hunyuan-Large (A52B) · 9.4
PIQAReasoning83.6%#14
best: GPT-4o mini · 93.1%
TriviaQAKnowledge71.5%#22
best: Sarvam-1 (2B) · 90.6%
WinoGrandeReasoning77.2%#48
best: PaLM 2 · 90.9%
Run it locally
VRAM @ Q4
26 GB
VRAM @ FP16
93 GB
Fits on (Q4)
RTX 5090 32GBM4 Pro 48GBM3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB
At Q4 (~26GB) a single RTX 4090 (24GB) needs CPU/GPU layer offloading; community reports of roughly 15-23 tok/s with partial offload
Quantizations
GGUF Q4 · AWQ · GPTQ
Mixtral family
Elo progression across releases
Mixtral 8x7BDec 2023438.9Mixtral 8x22BApr 2024678.2
Resources
HFmistralai/Mixtral-8x7B-v0.1GHmistralai/mistral-inferenceDOCSdocs.mistral.ai/models/model-cards/mixtral-8x7b-0-1APIAvailable via Mistral AI API
API price weights · each benchmark row carries its own source badge (see methodology)