Model components / Fine-tuned / chat models

Codestral

Mistral's proprietary code model, API-only via la Plateforme, Vertex AI, Azure AI Foundry, and Bedrock. Current SKU is Codestral 25.08 (August 2025), 30% more accepted completions and 50% fewer runaway generations versus 25.01. Part of Mistral's broader 2026 coding stack alongside Codestral Embed (retrieval-tuned embeddings), Devstral 2 (open-weight agentic coding counterpart), and Mistral Vibe CLI (Mistral's answer to Claude Code / Gemini CLI). Distributed to thousands of developers through Continue.dev, JetBrains, and VS Code integrations; reported usage trails Anthropic and OpenAI coding models.

Codestral 25.08 (released 30 Jul 2025), Mistral's code-completion/FIM specialist; instruction-tuned with a chat mode (+5% instruction following per launch). Distributed via API (console.mistral.ai), VPC, and on-prem; open weights on HF under the Mistral AI Non-Production License (MNPL). NOTE: arguably a coding-specialist model rather than a general chat model; scored best-effort in finetuned_chat per assignment with a category caveat. (Separate 'Codestral 2', Apr 2026, was relicensed Apache-2.0, not the SKU scored here.) Verified live June 2026.

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 downloadable but under MNPL (non-OSI, non-production); open-weights gated by a use-restricted license, so restricted not open_weights.

https://mistral.ai/news/codestral/ recorded 2026-06-04
Codestral open-weight on HF under Mistral AI Non-Production License (research/testing; commercial on demand)
https://mistral.ai/news/codestral-25-08/ recorded 2026-06-04
Codestral 25.08 released 30 Jul 2025; API + VPC + on-prem deployment

Adoption

3 low confidence

3.0

Embedded as the default FIM/code-completion backend in IDE assistants (Continue, Tabnine, and Mistral's own coding stack) and on multiple API gateways; Mistral reports per-version completion-acceptance gains but discloses no hard download/user count. Level 3 reflects multi-platform IDE distribution, not a verified usage_volume figure.

https://mistral.ai/news/codestral-25-08/ recorded 2026-06-04
Codestral coding stack distributed via API/VPC/on-prem; +30% accepted completions vs prior version

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.

https://mistral.ai/news/codestral/ recorded 2026-06-04
Mistral-reported HumanEval/MBPP/RepoBench/Spider code benchmarks for Codestral

Unchanged since 2026-06-24 (last edited, not re-checked)