Pixtral Large
Pixtral Large pairs a 123B decoder with a 1B vision encoder and 128K context; Mistral reports 69.4% on MathVista.
Built on Mistral Large 2, handles at least 30 high-res images per prompt and tops the LMSys vision leaderboard among open-weights models per Mistral. Released with an updated Large 24.11 text model. Research license for weights.
- Date
- Monday, 18 November 2024
- Lab
- Mistral AI
- Kind
- open-weights
- Access
- open weights (restricted license)
Figures
| Measure | Value | Measured by |
|---|---|---|
| MathVista | 69.4% | company |
Sources
This record was checked against its sources on 6 October 2026. How we check