Model components / Fine-tuned / chat models

Codestral

Mistral's code-completion and fill-in-the-middle specialist, instruction-tuned with a chat mode and reached through Mistral's API, a customer VPC or on-premises deployment. It sits in a broader coding stack alongside Codestral Embed for retrieval, Devstral as the open-weight agentic counterpart, and the Mistral Vibe CLI.

A coding specialist rather than a general chat model, scored here with that caveat. Mistral publishes this SKU's benchmark scores as chart images rather than text, so they are not re-derivable from the launch post. Verified 2026-08-13 via the Codestral and Codestral 25.08 launch posts.

Openness

2 high confidence

2.0

weights: open(downloadable on HF)
data: closed
code: closed(no training pipeline)
license: Mistral-AI-Non-Production-License(MNPL, non-OSI, research/testing only

Weights are downloadable, but only under the Mistral AI Non-Production License: not OSI-approved and limited to research and testing, with a commercial license available on request. Open weights behind a use-restricting license, so restricted rather than open weights.

https://mistral.ai/news/codestral/ recorded 2026-08-13
"Codestral is a 22B open-weight model licensed under the new Mistral AI Non-Production License, which means that you can use it for research and testing purposes. Codestral can be downloaded on HuggingFace ... Commercial licenses are also available on demand"
https://mistral.ai/news/codestral-25-08/ recorded 2026-08-13
Codestral 25.08 update; API, VPC and on-prem deployment for regulated environments

Adoption

3 low confidence

3.0

Embedded as the default fill-in-the-middle and code-completion backend in IDE assistants (Continue, Tabnine and Mistral's own coding stack) and on multiple API gateways. Mistral reports per-version gains in completion acceptance but discloses no hard download or user count, so level 3 reflects multi-platform IDE distribution rather than a verified figure.

https://mistral.ai/news/codestral-25-08/ recorded 2026-08-13
"+30% increase in accepted completions, +10% more retained code after suggestion, 50% fewer runaway generations"; distributed via API, VPC and on-prem; no download or user count published

Capability

3 medium confidence

3.0

Strong code-completion specialist for its size, but a narrow FIM model, mid-tier on a chat/agentic capability axis where the comparison set is 2026 frontier agentic coders. No SWE-bench figure for Codestral appears on any Mistral page, so the mid-tier agentic placement rests on the completion and FIM benchmarks rather than on an agentic score.

https://mistral.ai/news/codestral-2501 recorded 2026-08-14
Codestral 25.01 launch post, benchmarks published as HTML tables in the body. Codestral-2501 row - 256k context, HumanEval 86.6%, MBPP 80.2%, CruxEval 55.5%, LiveCodeBench 37.9%, RepoBench 38.0%, Spider 66.5%, CanItEdit 50.5%, HumanEval average 71.4%, HumanEvalFIM average 85.9%. Codestral-2405 22B row - 32k context, HumanEval 81.1%, MBPP 78.2%, RepoBench 34.0%, HumanEvalFIM average 82.1%. No MBPP 91.2% and no SWE-bench figure anywhere on the page.
https://mistral.ai/news/codestral/ recorded 2026-08-13
names the benchmark suite (HumanEval pass@1, MBPP sanitised pass@1, CruxEval, RepoBench EM, Spider, six-language HumanEval, FIM) and the claim that "Codestral outperforms all other models in RepoBench" with its 32k context; the scores themselves are chart images and are not in the fetched body

Verified 2026-08-13