Mistral 3: Mistral Large 3 and Ministral 3
Mistral Large 3, a 675B-parameter MoE with 41B active trained on 3,000 H200s, and Ministral 3 (3B/8B/14B) all ship under Apache 2.0.
Mistral's first open frontier-scale MoE and first fully Apache flagship since Mixtral. Multimodal and 40+ languages. Large 3 ranks #2 among open non-reasoning models on LMArena per Mistral; Ministral 14B reasoning reaches 85% on AIME 2025.
- Date
- Tuesday, 2 December 2025
- Lab
- Mistral AI
- Kind
- open-weights
- Access
- open weights
- Price
- Large 3 $0.50/$1.50 per 1M tokens, Dec 2025
Figures
| Measure | Value | Measured by |
|---|---|---|
| Active / total parameters (Large 3) | 41B / 675B MoE, trained on 3,000 Nvidia H200 GPUs | company |
| AIME 2025 (Ministral 14B reasoning) | 85% | company |
| API price (Large 3) | $0.50 in / $1.50 out per 1M tokens | company |
LMArena rank is Mistral's report; 'as of Aug 2026' Large 3 was still Mistral's largest public model per a secondary roundup.
Sources
This record was checked against its sources on 6 October 2026. How we check