Mistral Large is the top tier of Mistral AI's general models; the newest, Mistral Large 4, is a multimodal model with 1 trillion parameters and open weights.
In plain words
The first Mistral Large, in February 2024, was the company's flagship text model offered through its API. Mistral Large 2, in July 2024, had 123 billion parameters under a research-only licence. Mistral Large 3, released in December 2025, was a mixture-of-experts model with 675 billion parameters, 41 billion active per token, under Apache 2.0. Mistral Large 4, nicknamed Le Chonk, came next: a 1-trillion-parameter multimodal model, first offered as a preview and then released with open weights under a modified MIT licence.
A quick example
Imagine a company that wants a strong model it can host itself. It could download Mistral Large 4's weights from Hugging Face, run them on its own servers with vLLM and fine-tune them. Selling the model as a hosted service would need a separate licence from Mistral.
Where you'll meet it
- Mistral's API, la Plateforme.
- Self-hosted deployments using the open weights.
Related
Sources
- DocsAu Large, Mistral AI · read 12 Oct 2026
- DocsLarge Enough, Mistral AI · read 29 Sept 2026
- DocsIntroducing Mistral 3, Mistral AI · read 12 Oct 2026
- DocsModels overview, Mistral AI docs · read 29 Sept 2026
- DocsMistral Large 4 (Le Chonk) is now open weight, Mistral AI · read 29 Oct 2026
Checked by WhatIsAI against the sources above · 30 Oct 2026. Spotted an error? Report it.