Term
Codestral
Codestral is Mistral AI's code-specialized model (first released May 2024) — a 22B-parameter model for code completion and generation, originally published as open weights.
Codestral — explained in more detail
Codestral is Mistral AI’s programming-specialized model, first introduced in May 2024. Unlike generalist chat models, it is trained specifically for code tasks — code completion, generating whole functions, tests and fill-in-the-middle across many programming languages. The first version (Codestral 2405) was a 22-billion-parameter model with open weights under Mistral’s Non-Production License (MNPL), which permits research and testing. The original context window was 32K tokens.
Since then Mistral has evolved the line across several versions. The current generation (Codestral 2508 / 25.08, released in late July / August 2025) extends the context window to 256,000 tokens and reports roughly 30% more accepted completions and 50% fewer runaway generations.
Example / Practical use
Codestral targets developer tooling specifically: IDE autocomplete, code assistants and CI integrations. The fill-in-the-middle capability is useful because editors know both the text before and after the cursor position while typing. With the 256K context, larger files or several related modules can be considered in a single call. Access is via Mistral’s API and coding stack; the open weights of the early version also allow local deployment for testing and research purposes.
Distinction
Codestral is not a general-purpose chat model like Mistral Large but a specialist for code. Within Mistral’s lineup it is the tool for developer workflows, while the generalist models cover broad knowledge and dialogue tasks. Note the licensing: the open weights are under the Non-Production License, while production use runs through the commercial API.
Discover more
GPT-6 Astra Found Questions My First Security Review Missed
I used GPT-6 Astra and Fable 5.1 as independent reviewers for RLS, API and tenant-isolation checks. The useful part was the structured cross-review.
GlossaryMagistral
Magistral is Mistral AIs first reasoning model from June 2025 — with a transparent, verifiable chain of thought and strength in European languages, partly as open weights.
EncyclopediaComparing AI models — who builds what and how to choose
The major model families in 2026 at a glance. Who builds Claude, GPT, Gemini, Llama, Mistral, DeepSeek, Qwen — and which model to pick when.