Codestral is Mistral's code-specialized large language model for developers who want fast, context-aware completions and fill-in-the-middle edits across more than 80 programming languages. It sits between a general chatbot and a full coding agent: you point it at a file or a function and it returns working code, not a conversation.
Core Features
- Training on 80+ programming languages, with strong coverage of Python, JavaScript, TypeScript, Go, C++, and Rust.
- A 32,000-token context window sized for whole-file and multi-file edits.
- Fill-in-the-middle support, so it can complete a span given the code before and after the cursor.
- A dedicated endpoint at codestral.mistral.ai with token-based billing separate from the general Mistral chat plans.
- Native IDE plugins including Continue.dev and Tabnine, so it runs inside VS Code and JetBrains without leaving the editor.
Use Cases
- Individual developers who want inline completions that respect the surrounding file.
- Teams standardizing on a self-hostable code model behind their own keys.
- Platforms like Tabnine that embed a specialized model instead of a general LLM.
Pricing
Codestral is available through Mistral's Codestral API on a token-based plan, and it is free to use inside IDE plugins such as Continue.dev and Tabnine for individual developers. Production and team usage moves to the paid API tier, which separates code-model cost from general Mistral chat subscriptions.
Our Take
Best for shops that want a focused, fast code model with clean IDE integration and no chatbot overhead. The trade-off is that Codestral writes and edits code but does not plan multi-file refactors or run tests, so pair it with Claude Code or Codex for full agentic workflows. See related AI Coding tools.
