# Codestral Codestral is a 22B parameter code generation model developed by [[Mistral AI]], released in May 2024. It is Mistral's first model explicitly designed for code generation tasks, trained on a diverse dataset of 80+ programming languages including Python, Java, C, C++, JavaScript, Bash, Swift, and Fortran. The model supports both instruction and fill-in-the-middle (FIM) completion through a shared API endpoint. It can complete coding functions, write tests, and fill in partial code. With a 32K token context window (larger than competitors' 4K-16K), Codestral outperformed all other models on RepoBench for long-range repository-level code completion at the time of release. Codestral is integrated into popular developer tools including Continue.dev and Tabnine for VSCode and JetBrains environments, as well as application frameworks like LlamaIndex and LangChain. It is available via La Plateforme API and a dedicated `codestral.mistral.ai` endpoint. The model was released under the Mistral AI Non-Production License (MNPL) for research and testing, with commercial licenses available on request. ## References - https://mistral.ai/news/codestral ## Related - [[Mistral AI]] - [[Large Language Models (LLMs)]] - [[Mistral Small 4]]