Overview
Codestral 25.08 (API id codestral-2508) is Mistral AI's flagship code-generation model, released on July 30, 2025 as the centerpiece of the company's enterprise "Mistral Coding Stack" (alongside the Devstral coding agents, the Codestral Embed retrieval model, and the Mistral Code IDE extension for VS Code and JetBrains). It specializes in low-latency, high-frequency developer tasks such as code completion, fill-in-the-middle (FIM), code correction, and test generation across 80+ programming languages.
Compared with the previous Codestral 25.01 release, Mistral reports that Codestral 25.08 delivers a 30% increase in accepted code completions, roughly 10% more retained code after a suggestion is accepted, and 50% fewer "runaway" generations, plus about a 5% gain on instruction-following (IF eval) and a 5% improvement in average MultiplE code ability. The model exposes a 256K-token context window with up to roughly 16K output tokens, and supports chat completions, function calling, structured outputs, and dedicated FIM completions through the Mistral API.
Unlike the original open-weight Codestral 22B (24.05), Codestral 25.08 is a proprietary "Premier" model: it is available through the Mistral API (and resellers such as OpenRouter) rather than as downloadable weights, though Mistral offers it for cloud, VPC, and on-premise enterprise deployment with no architectural changes. Mistral lists it at $0.30 per million input tokens and $0.90 per million output tokens.
| Released | 2025-07-30 |
|---|---|
| License | Proprietary (Mistral commercial / Premier; API-only, not open weights) |
| Weights | API only |
| Parameters | Not publicly disclosed |
| Context | 256K |
| Max output | 16K tokens |
| Architecture | Decoder-only Transformer optimized for low-latency code completion and fill-in-the-middle; supports chat and instruction following. Parameter count not publicly disclosed. |
| Knowledge cutoff | Not disclosed |
| Modalities | Text |
| Status | Generally available |
Pricing
| Input | $0.30 / 1M tokens per 1M tokens |
|---|---|
| Output | $0.90 / 1M tokens per 1M tokens |
Mistral API list price for the Premier 'Codestral' model; same rates shown on OpenRouter.
Strengths
- Low-latency, high-frequency code completion and fill-in-the-middle (FIM) tuned for real-time IDE autocomplete
- Reports 30% more accepted completions, ~10% more retained code, and 50% fewer runaway generations versus Codestral 25.01
- Large 256K-token context window for reasoning over big files and multi-file repositories
- Broad language coverage (80+ programming languages) plus chat, function calling, and structured outputs
- Inexpensive at $0.30 / $0.90 per million input/output tokens, with flexible cloud, VPC, and on-prem enterprise deployment
Best for
- Real-time IDE code completion and fill-in-the-middle inside Mistral Code (VS Code / JetBrains) or other editors
- Code generation, refactoring, and bug fixing across many programming languages
- Automated test generation and code correction
- Self-hosted or VPC coding assistants for regulated enterprises (banking, defense, energy)
How to access
| Provider | Model ID |
|---|---|
| Mistral AI ↗ | codestral-2508 |
| OpenRouter ↗ | mistralai/codestral-2508 |
Codestral — every version
The full lineage of the Codestral line, newest first. Every version has its own page — click any to compare specs, benchmarks and pricing.
| Version | Released | Context | License |
|---|---|---|---|
| Codestral (25.08)current | 2025-07-30 | — | Open weights |
| Codestral 25.01 | 2025-01 | — | Open weights |
| Codestral 22B (24.05) | 2024-05-29 | — | Open weights |
FAQ
What is Codestral 25.08?
Codestral 25.08 (API id codestral-2508) is Mistral AI's flagship code-generation model, released on July 30, 2025. It is tuned for low-latency tasks like code completion, fill-in-the-middle, code correction, and test generation across 80+ programming languages, and is the core model of Mistral's enterprise Coding Stack.
Is Codestral 25.08 open weights?
No. Unlike the original open-weight Codestral 22B (24.05), Codestral 25.08 is a proprietary Premier model available through the Mistral API and resellers such as OpenRouter, not as downloadable weights. Mistral does offer it for cloud, VPC, and on-premise enterprise deployment.
How much does Codestral 25.08 cost?
Mistral lists Codestral at $0.30 per million input tokens and $0.90 per million output tokens, and OpenRouter shows the same rates.
How is Codestral 25.08 better than Codestral 25.01?
Mistral reports that Codestral 25.08 produces about 30% more accepted code completions, retains roughly 10% more accepted code, and has 50% fewer runaway generations, plus around a 5% gain on instruction following (IF eval) and a 5% improvement in average MultiplE code ability.