Side-by-side comparison of Codestral 22B (Mistral AI · France) and DeepSeek V3.2 (DeepSeek · China) for self-hosted deployment of the open-weight model. Codestral 22B is rated conditional; DeepSeek V3.2 is conditional. They part ways on licence: Codestral 22B is "MNPL (non-prod)", DeepSeek V3.2 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 685B successor to V3 with DeepSeek Sparse Attention for long context, scalable RL for agentic tasks. Vendor claims parity with GPT-5 (Speciale variant exceeds). MIT licence keeps weights clean; Chinese-origin considerations unchanged. |
| Last reviewed | 2026-04-15 | 2026-04-15 |
| Open-weight | ||
| Licence | MNPL (non-prod) | MIT |
| Commercial use | Paid licence req. | Yes |
| Training data | Partial | Undisclosed |
| Origin | EU | China |
| Performance & pricing? | ||
| Quality index | — | 32/100 |
| Speed | — | 32 tok/s |
| Blended price | — | $0.32/M |
| Context window | — | — |
| Evidence | ||
| Sources | ||
No overlapping sources between the two entries.