Side-by-side comparison of Codestral 22B (Mistral AI · France) and Llama 3.3 70B (Meta · USA) for self-hosted deployment of the open-weight model. Codestral 22B is rated conditional; Llama 3.3 70B is conditional. They part ways on licence: Codestral 22B is "MNPL (non-prod)", Llama 3.3 70B is "Llama community".
| 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 Strong 70B model, near-flagship quality at smaller size. Same Llama community licence as Llama 4: 700M MAU cap, acceptable-use policy, 'Built with Llama' attribution required. |
| Last reviewed | 2026-04-15 | 2026-04-15 |
| Open-weight | ||
| Licence | MNPL (non-prod) | Llama community |
| Commercial use | Paid licence req. | With caps |
| Training data | Partial | Undisclosed |
| Origin | EU | USA |
| Performance & pricing? | ||
| Quality index | — | 15/100 |
| Speed | — | 97 tok/s |
| Blended price | — | $0.68/M |
| Context window | — | — |
| Evidence | ||
| Sources | ||
No overlapping sources between the two entries.