Mistral Large 4 arrives as a paid preview, with open weights promised by the end of October
A one-trillion-parameter multimodal model from Mistral is available through its API now. The downloadable weights, and their licence, are still to come.
Mistral AI announced Large 4 on 6 October 2026. It describes a natively multimodal model of about one trillion parameters, with 52 billion active at a time, that understands images and documents as well as text. The Next Web gives the active figure as 49 billion; Mistral’s own post says 52 billion.
For now it is a public preview through Mistral’s API, priced at $1.36 per million input tokens and $4.18 per million output tokens on its product card. Mistral calls the model open-weight and says it will release the weights by the end of the month. The Next Web reports a date of 27 October. Mistral has named no licence, so it is not yet clear what self-hosting will be allowed.
The benchmark results in the announcement are Mistral’s own and are described as preliminary. They include 61.7% on DeepSWE v1.1 and a top-five place on the Artificial Analysis Cyber Index. Independent scores will matter more once the weights are public.
Mistral says the model was trained on about 4,000 Nvidia Grace Blackwell GPUs. It has not published what it takes to run. As a rough guide, one trillion parameters at 4 bits each is about 500 GB for the weights alone, which is server hardware, not a desktop.
Sources
This is a summary of reporting published elsewhere. Where a claim matters, follow the source — it is the record, this is not.
Related on this site
Which models can your PC run?
Scan your hardware and see every model that fits, with the memory each one needs.