Codestral 25.01
Codestral 25.01 is about 2x faster than its predecessor, with a 256K context and 95.3% fill-in-the-middle pass@1.
Better tokenizer and efficiency for completion in IDEs, with HumanEval at 71.4% (from 65.6%) and context from 32K to 256K. Debuted #1 on the LMSys Copilot Arena leaderboard per Mistral.
- Date
- Monday, 13 January 2025
- Lab
- Mistral AI
- Kind
- model
- Access
- closed API
Figures
| Measure | Value | Measured by |
|---|---|---|
| HumanEval | 71.4% from 65.6% | company |
| Fill-in-the-middle pass@1 | 95.3% | company |
Sources
This record was checked against its sources on 6 October 2026. How we check