Codestral 25.06
Coding-specialized open model.
Modality
Code
Context
Code-optimized context
Pricing (per 1M)
Open / self-host
What Codestral 25.06 is for
Codestral is a code specialist rather than a general model that also writes code, and that distinction is the entire argument for using it.
Specialist models trade breadth for depth. Codestral will not help you draft an email, but for completion, generation and translation between languages it delivers more per dollar than a general flagship doing the same job.
Where it does well
Fill-in-the-middle completion is where specialists earn their place. Understanding code before and after the cursor — not just continuing from a prefix — is what makes editor integration feel useful rather than intrusive, and general models are weaker at it.
Latency and cost make it viable for the loop that matters: a model that suggests as you type has to be fast enough not to interrupt thought.
Where it falls short
It is not a reasoning model. Architecture decisions, debugging that requires understanding intent, and code review that weighs trade-offs are all jobs for a general flagship.
Non-code tasks — documentation prose, commit messages with context, explaining a decision to a non-engineer — are outside what it is built for.
Should you use it?
Use Codestral in the editor for completion and generation, and a general flagship for review, architecture and debugging. That split is cheaper and better than either model doing both — the specialist handles volume, the generalist handles judgement.
Capabilities
Pros
- Strong at code for its size
- Open + self-hostable
- Fast
Cons
- Narrow: built for code, not general chat
Best for
Try a prompt for Codestral 25.06
Complete this function and add docstrings and type hints.
Other Mistral models
Pricing indicative, per 1M tokens, reviewed July 2026.
Visit official site