Side-by-side comparison of Codestral 22B (Mistral AI · France) and DeepSeek-V4-Pro (DeepSeek · China) for self-hosted deployment of the open-weight model. Codestral 22B is rated conditional; DeepSeek-V4-Pro is conditional. They part ways on licence: Codestral 22B is "MNPL (non-prod)", DeepSeek-V4-Pro is "MIT".
| Field | ||
|---|---|---|
| Summary | ||
| Verdict | Conditional EU code model trained on 80+ languages. Licensed under Mistral Non-Production License — blocked for any production or commercial deployment without a paid commercial licence. Use Codestral Mamba (Apache 2.0) if you need commercial freedom. | Conditional Per the published LICENSE file, DeepSeek-V4-Pro ships under MIT with no commercial restrictions, so the weights themselves are deployable. The caveats are non-EU jurisdiction and a training corpus described only by aggregate token count (32T+) without a dataset list — both should be documented in any GDPR or AI Act compliance file before regulated use. |
| Last reviewed | 2026-04-15 | 2026-04-28 |
| Open-weight | ||
| Licence | MNPL (non-prod) | MIT |
| Commercial use | Paid licence req. | Unrestricted |
| Training data | Partial | Categories only |
| Origin | EU | China |
| Performance & pricing? | ||
| Quality index | — | 52/100 |
| Speed | — | 36 tok/s |
| Blended price | — | $2.17/M |
| Context window | — | — |
| Evidence | ||
| Sources | ||
No overlapping sources between the two entries.