Codestral

Summary

Codestral is Mistral AI's code-specialized model — purpose-built for software engineering tasks including code generation, completion, debugging, and explanation. It is the only Mistral model released under a non-Apache license. Codestral is available via Mistral's API and has been integrated into popular developer tools including VS Code and JetBrains IDEs.

Overview

Codestral is Mistral AI's code-specialized model — purpose-built for software engineering tasks including code generation, completion, debugging, and explanation. It is the only Mistral model released under a non-Apache license (the Mistral AI Non-Production License for free use; a commercial license is required for production), reflecting the higher commercial value of specialized coding models. Codestral is available via Mistral's API and has been integrated into popular developer tools including VS Code (via Continue and other extensions) and JetBrains IDEs.

For developers who are already in the Mistral ecosystem or who prefer European-hosted AI infrastructure, Codestral is a strong specialized alternative to Anthropic's Claude Sonnet and OpenAI's GPT-5.2 Codex for coding tasks.

Specifications

  • Developer: Mistral AI
  • Model String: codestral-latest (check Mistral docs for versioned string)
  • Release Date: May 2024 (original); updated versions released throughout 2025
  • Type: Code-specialized Large Language Model
  • Context Window: 256,000 tokens
  • Languages: 80+ programming languages
  • License: Mistral AI Non-Production License (free non-commercial); commercial license required for production
  • Access: Mistral API (la Plateforme), VS Code extensions, JetBrains integrations, self-hosted (non-commercial)
  • Pricing: See Mistral pricing for current rates

Capabilities

80+ Programming Languages: One of the broadest language coverage ranges of any code model — handles mainstream languages (Python, JavaScript, Java, C++) as well as less common ones.

Fill-in-the-Middle (FIM): Supports fill-in-the-middle completion — given code before and after a cursor position, it generates the code that belongs in between. Essential for inline IDE code completion workflows.

256K Context: Can hold large codebases in context, enabling cross-file analysis and understanding of complex project structures.

IDE Integration: First-class support in popular developer environments via extensions, making it practical for day-to-day coding workflows rather than just API access.

Code Explanation & Debugging: Beyond generation, strong at explaining what existing code does and identifying bugs.

Limitations

Mistral's Non-Production License is a meaningful restriction for commercial deployment — developers must obtain a paid commercial license before using Codestral in production. On the most demanding coding benchmarks (SWE-bench Verified), it trails Anthropic's Claude Sonnet 4.6 and OpenAI's GPT-5.2 Codex.

Recent Developments

  • 2025 Updates: Codestral has received iterative improvements throughout 2025, with context window expansion to 256K and improved fill-in-the-middle performance.
  • Devstral Variants: Mistral released Devstral Small and Medium variants in 2025, extending the code-specialized family to smaller, more deployable model sizes.

Last Updated

February 26, 2026