Model

Mistral Large 2 (24.07)

OPEN

Mistral AI · Mistral Large family · released Jul 24, 2024

123B open-weights flagship rivaling Llama 3.1 405B on code/math (research license); fully retired from Mistral's API as of 2025.

ReasoningCodingVisionFunction callingTool useAgentic

1068.2

Elo · rank #200

Parameters

123B

Active params

123B (dense)

Context

128K tokens

Architecture

Dense Transformer

License

Mistral Research License (non-commercial; commercial license required for production)

Languages

13+

API price (in/out)

No hosted API

Modalities

text

Benchmark results

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

Arena-HardHuman preference73.2%#33

best: Qwen3-235B-A22B (Non-Thinking) · 96.1%

GSM8KMath93.0%#23

best: Llama 3.1 405B · 96.8%

HumanEval+Coding87.0%#8

best: Mistral Small 3.2 (24B) · 92.9%

HumanEvalCoding92.0%#15

best: Claude Opus 4.5 · 99.4%

LogicKorHuman preference9.03#6

best: GPT-4o · 9.33

MATH-500Math70.0%#91

best: GPT-5 · 99.4%

MBPP+Coding69.0%#17

best: Llama 3.1 405B · 88.6%

MBPPCoding80.0%#15

best: Llama-3.3-Nemotron-Super-49B v1 (Reasoning On) · 91.3%

MMLU-ProKnowledge65.9%#104

best: Claude Fable 5 · 91.5%

MMLUKnowledge84.0%#68

best: OpenAI o3 · 92.9%

MT-BenchHuman preference8.63#20

best: Hunyuan-Large (A52B) · 9.4

WildBenchHuman preference56.3%#3

best: Mistral Small 3.2 (24B) · 65.3%

Run it locally

VRAM @ Q4

74 GB

VRAM @ FP16

246 GB

Fits on (Q4)

M3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB

123B model exceeds a single RTX 4090's 24GB VRAM even at Q4; requires multi-GPU or CPU offload.

Quantizations

GGUF Q4 · AWQ · MLX

Mistral Large family

Elo progression across releases

Mistral LargeFeb 2024713.0 Mistral Large 2 (24.07)Jul 20241068.2 Mistral Large 24.11Nov 2024658.0 Mistral Large 3Dec 20251303.2

Resources

HFmistralai/Mistral-Large-Instruct-2407 DOCSdocs.mistral.ai/models/overview

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