HomeWinBuzzer NewsMistral AI Launches Codestral-22B for Enhanced Code Generation

Mistral AI Launches Codestral-22B for Enhanced Code Generation

Codestral-22B brings Mistral AI coding capabilities to over 80 programming languages, including Python, Java, C++, and JavaScript.


has introduced its latest artificial intelligence model, Codestral-22B, designed to support over 80 languages, including Python, Java, C++, and JavaScript. This model aims to streamline software coding development by reducing errors and improving efficiency for developers.

Codestral-22B is an open-weight generative AI model that caters to a diverse range of programming languages. From widely-used languages like Python and Java to more specialized ones such as Swift and Fortran, the model's extensive language support makes it a versatile tool for various coding environments. It assists developers by completing coding functions, writing tests, and filling in partial code, thereby minimizing the risk of errors.

Performance and Integration

The model's performance is noteworthy, featuring a 32k context window that excels in long-range evaluations like RepoBench. Codestral-22B has demonstrated superior performance across multiple benchmarks, including HumanEval pass@1 for Python, the Spider benchmark for SQL, and evaluations for languages such as C++, Bash, Java, PHP, Typescript, and C#.

Codestral outperforms all other models in RepoBench

It is available for download under the Mistral AI Non-Production License for research and testing purposes and can be accessed via HuggingFace. Additionally, a dedicated endpoint, codestral.mistral.ai, optimized for IDE integrations, is available through a personal API key, free during an 8-week beta period managed via a waitlist. The open sourcing of Codestral is important as Mistral has said it is committed to open source AI despite a recent major investment partnership from Microsoft

Model Capabilities

Beyond supporting over 80 programming languages, Codestral-22B can write software in low-level syntaxes, allowing programmers to interact more directly with underlying hardware. This feature is particularly useful for optimizing program performance. The model can also explain code snippets, generate new code based on natural language instructions, and function as an autocomplete tool for partially written code. Additionally, it can modify existing code components, such as changing specific lines within a snippet.

Codestral-22B can automatically scan code for flaws, speeding up the process of testing newly created software for bugs. This includes detecting issues that may not be immediately apparent through manual troubleshooting. In internal evaluations, Codestral-22B outperformed three other open-source LLMs, including Meta Platforms Inc.'s Llama 3 70B, particularly in Python programming tasks. It achieved the second-highest score in SQL evaluations.

AI Coding Open Source Battlefield

One of the hardest fought parts of the AI industry is coding tools. We have seen many models released, from major companies to smaller startups. For example, Codestral-22B is not the only AI coding tool to launch this week. The JACoB AI also made its debut this week as an open source automated coding solution. 

JACoB works smoothly alongside the tools you already use, like GitHub and others. It's a multi-talent: writing code, reviewing what you've written, and even turning designs from Figma into real code. Plus, it automates tedious tasks like pull requests and keeps an eye out for security problems and bugs. The more you use JACoB, the better it gets at understanding your coding style and generating high-quality results.

Luke Jones
Luke Jones
Luke has been writing about all things tech for more than five years. He is following Microsoft closely to bring you the latest news about Windows, Office, Azure, Skype, HoloLens and all the rest of their products.

Recent News