Mistral-Large-4-1T-A52B: Chat, code and ask about images (1T, 52B active)

Quick questions

What is Mistral-Large-4-1T-A52B?

Mistral-Large-4-1T-A52B is a Mixture-of-Experts model, meaning it is split into many specialist sub-networks and only a few run for each word.

What can I use Mistral-Large-4-1T-A52B for?

Upload a photo of a technical drawing or chart and ask a question about it, and you get a written answer describing what the image shows; Send a request with a strict JSON schema, meaning a template listing required fields, and you get a reply that fills exactly those fields, ready for your app; Write a question in Polish or Japanese, and you get an answer in that language; the model card lists fluency in 160+ languages

Can I run Mistral-Large-4-1T-A52B on my computer?

At about 1 trillion parameters in total (52 billion active), it needs a multi-GPU server or the hosted API, not a desktop or laptop. If GPU memory runs short, lower the context length or number of parallel requests.

What do I need to use Mistral-Large-4-1T-A52B?

The recommended vLLM Docker setup uses NVIDIA GPUs; exact GPU memory needs are not documented. A quantized NVFP4 version with a smaller memory footprint targets Blackwell GPUs.

What license does Mistral-Large-4-1T-A52B use?

Licensed under the Mistral Ecosystem Open-Weight License: commercial and non-commercial use are allowed, but running a Model-as-a-Service business needs a separate commercial license from Mistral. License text: https://huggingface.co/mistralai/Mistral-Large-4-1T-A52B/blob/main/LICENSE.