Mistral introduces Codestral, its inaugural AI model designed for programming tasks

Paris-based AI startup Mistral has made a groundbreaking entry into the programming and development space with the launch of Codestral, its first-ever code-centric large language model (LLM). This 22B parameter, open-weight generative AI model specializes in coding tasks, covering more than 80 programming languages and outperforming previous models like CodeLlama 70B and Deepseek Coder 33B.

Codestral, available under a non-commercial license, aims to revolutionize the coding experience for software developers by providing advanced AI applications. Mistral claims that the model can help developers write and interact with code in various coding environments and projects, saving time and effort while reducing the risk of errors and bugs.

In recent tests, Codestral has shown impressive performance across different programming languages, outperforming competitors on various evaluation platforms like RepoBench and HumanEval. Popular tools for developer productivity and AI application development, including LlamaIndex, LangChain, and JetBrains, have already started testing Codestral and have praised its capabilities.

Interested developers can access Codestral on Hugging Face under a non-production license for non-commercial purposes and research work. The model is also available via two API endpoints for different use cases. Mistral’s move to introduce Codestral provides enterprise researchers with a powerful tool to accelerate software development, competing with other code-centric models in the market like StarCoder2, OpenAI’s Codex, and Amazon’s CodeWhisper.

As the AI Impact Tour approaches on June 5th in NYC, developers have a limited time to request an invite and explore the latest methods for auditing AI models. Don’t miss out on this exclusive opportunity to stay ahead in the rapidly evolving AI landscape.