Mistral Large 3
Mistral Large 3 is an open-weight language model from Mistral AI, released on December 2, 2025.
- Developer
- Mistral AI
- Released
- December 2, 2025
- Total parameters
- 675B
- Active parameters per token
- 41B
- Architecture
- MoE
- Context window
- 256K
- Input
- Text, image
- License
- Apache 2.0
- Commercial use
- Yes
- Hugging Face repository
- huggingface.co
- Checked on
Mistral Large 3 is a mixture-of-experts model with 675B total parameters, of which 41B are active for each token. Mistral AI lists a context window of 256K tokens. The model accepts text and image input. The weights are released under Apache 2.0, which allows commercial use.
- Total parameters, billions
- 675B
- Active parameters, billions
- 41B
- Context window, tokens
- 256K
- License type
- Apache 2.0
- License conditions
- Apache 2.0, with no user or revenue limits.
- Weights precision
- FP8
- Official quantized or alternative versions
- mistralai/Mistral-Large-3-675B-Instruct-2512-NVFP4; mistralai/Mistral-Large-3-675B-Instruct-2512-BF16 (full precision); mistralai/Mistral-Large-3-675B-Instruct-2512-Eagle
- License text
- apache.org
- Sources
Page URL Hugging Face model card https://huggingface.co/mistralai/Mistral-Large-3-675B-Instruct-2512/raw/main/README.md mistral.ai announcement https://mistral.ai/news/mistral-3
Questions
What license does Mistral Large 3 use?
The weights are released under Apache 2.0, which allows commercial use with no user or revenue limits.
How much memory does Mistral Large 3 need?
Mistral Large 3 has 675B total parameters, and all of them must be loaded. The main weights are published in FP8, with NVFP4 and full-precision BF16 versions also available; the memory calculator checks which setups fit.
How many parameters does Mistral Large 3 have?
675B in total, of which 41B are active for each token, with a context window of 256K tokens.
What input does Mistral Large 3 accept?
Text and image input.
Sources
Every figure comes from the developer’s model card, license text and official announcements, and from the model’s Hugging Face repository metadata. Each model page lists its sources and the date they were checked; the register is rechecked every week.
