Model

Codestral 25.01

CLOSED

Mistral AI · Codestral family · released Jan 13, 2025

Faster closed Codestral refresh: improved tokenizer, ~2x speed, 256k context; debuted #1 on LMSYS Copilot Arena.

ReasoningCodingVisionFunction callingTool useAgentic

1056.8

Elo · unrated

Parameters

Undisclosed

Active params

—

Context

256K tokens

Architecture

Undisclosed dense transformer, improved tokenizer vs Codestral 22B (closed weights)

License

Proprietary (Mistral Commercial License)

Languages

—

API price (in/out)

$0.3 / $0.9

Modalities

text

Benchmark results

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

Aider PolyglotCoding11.1%#38

best: Claude Opus 4.5 · 89.4%

HumanEval FIMCoding85.9%#3

best: Codestral 22B · 91.6%

HumanEvalCoding86.6%#38

best: Claude Opus 4.5 · 99.4%

LiveCodeBenchCoding37.9%#135

best: DeepSeek-V4-Pro (Think Max) · 93.5%

MBPPCoding80.2%#14

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

RepoBenchCoding38.0%#5

best: DeepSeek Coder 33B · 41.9%

Run it locally

Closed weights — available via API only. No local deployment.

Input / M tok

$0.3

Output / M tok

$0.9

Codestral family

Elo progression across releases

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

Resources

DOCSdocs.mistral.ai/models/codestral-25-01

API price $0.3/$0.9 · each benchmark row carries its own source badge (see methodology)