AI BriefingCodestral 25.01
Key point
Mistral AI unveiled Codestral 25.01, which improves the architecture and tokenizer to double the speed.
Details
Mistral AI's coding model Codestral has been upgraded to Codestral 25.01 through a major update.
This version features a more efficient architecture and an improved tokenizer, making code generation and completion about 2x faster than before. It currently leads coding models in the same weight class, and in particular achieves SOTA (State-of-the-art) performance on FIM (Fill-In-the-Middle) tasks.
Key benchmark results are as follows:
- HumanEval: 86.6%
- MBPP: 80.2%
- CruxEval: 55.5%
- LiveCodeBench: 37.9%
These results surpass major competing models such as the existing Codestral 2405 22B and DeepSeek Coder 33B.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.