Mistral AI released Codestral 25.08, its latest flagship open coding model. Featuring native 256k context and low-latency Fill-in-the-Middle completion across 80+ programming languages, it outperforms peers in its size class by over 15% on MultiPL-E and HumanEval-Plus benchmarks.
Key Takeaways
- βNative 256k context window ingests entire repository trees for accurate symbol resolution.
- βEnhanced Fill-in-the-Middle (FIM) optimization provides sub-second multi-line autocomplete.
- βWeights live on Hugging Face with hosted API endpoints and instant Ollama setup.