Model

Codestral 22B

OPEN

Mistral AI · Codestral family · released May 29, 2024

Mistral's first dedicated code model; fill-in-the-middle across 80+ programming languages, 32K context, non-commercial MNPL license.

ReasoningCodingVisionFunction callingTool useAgentic

458.9

Elo · rank #317

Parameters

22.2B

Active params

22.2B (dense)

Context

32K tokens

Architecture

Dense transformer (code-specialized, fill-in-the-middle)

License

Mistral AI Non-Production License (MNPL-0.1)

Languages

—

API price (in/out)

No hosted API

Modalities

text

Benchmark results

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

BIG-Bench HardReasoning30.7%#131

best: ERNIE 4.5 300B-A47B · 94.3%

GPQA DiamondReasoning6.5%#287

best: GPT-5.6 · 94.6%

HumanEval FIMCoding91.6%#1

best: this model · 91.6%

HumanEvalCoding81.1%#53

best: Claude Opus 4.5 · 99.4%

IFEvalReasoning57.7%#124

best: Gemma 4 26B A4B · 98.5%

MBPPCoding78.2%#18

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

MMLU-ProKnowledge23.9%#168

best: Claude Fable 5 · 91.5%

RepoBenchCoding34.0%#8

best: DeepSeek Coder 33B · 41.9%

Run it locally

VRAM @ Q4

18 GB

VRAM @ FP16

44 GB

Fits on (Q4)

RTX 3090 24GBRTX 4090 24GBRTX 5090 32GBM4 Pro 48GBM3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB

Fits a single RTX 4090 at Q4_K_M (~18GB); no single authoritative tok/s benchmark found

Quantizations

GGUF Q4 · AWQ · GPTQ

Codestral family

Elo progression across releases

Codestral 22BMay 2024458.9 Codestral Mamba (7B)Jul 2024774.0 Codestral 25.01Jan 20251056.8

Resources

HFmistralai/Codestral-22B-v0.1 GHmistralai/mistral-inference DOCSlegal.mistral.ai/ai-governance/models/codestral APIAvailable via Mistral AI API

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