Codestral

Codestral is Mistral AI's model for writing code, and specifically for the part of writing code that a chat window handles badly: the completion that appears while you type. Mistral describes it as specialised in low-latency, high-frequency tasks such as fill-in-the-middle and code generation. Fill-in-the-middle is the technical name for the thing that matters here — the model is given the code before the cursor and the code after it, and has to produce something that fits between them. That is a different job from answering a question about code, and it is the job an editor plug-in needs done thousands of times a day. The current version carries the v25.08 tag. Unlike much of Mistral's line-up, Codestral is not open-weight: it is a commercial model available through Mistral's platform, priced at $0.30 per million input tokens and $0.90 per million output tokens. That is roughly a fifth of what Mistral Medium 3.5 costs on input, which is the economics an autocomplete has to have — a model invoked on every keystroke pause cannot be priced like a model invoked once per question. Mistral does not publish a context window, a supported-language count or benchmark figures for this version on the documentation pages we could read, so this profile does not state them; we will fill them in if a model card appears. The catalogue lists Codestral separately from Mistral's general-purpose models because it is sold and versioned separately, on its own release cadence.

▸Videos

No videos yet.

▸Reviews