Back to the stories

Mistral AI launches public preview of 1-trillion-parameter multimodal frontier model Mistral Large 4 "Le Chonk"

Score 10.0,

LLMs

Developers can now try a frontier-scale multimodal model in a public preview.

Mistral calls it Mistral Large 4, nicknamed "Le Chonk," and has opened access via the Mistral Studio API. The company describes ML4 as an open-weight, natively multimodal flagship built to combine instruction following, deeper reasoning and agent-style planning in one model.

What makes it stand out is scale and design: Mistral says ML4 works at roughly one trillion parameters and uses a Mixture-of-Experts architecture to be huge without being wasteful. The preview gives developers API-level access now, while Mistral plans to publish the model weights later in October after safety testing.

Why this matters: an open-weight frontier model at this scale changes the playground. Until now, the most capable systems were mostly closed behind a few large providers. Early API access, and a promised weight release, mean more teams can experiment directly with frontier capabilities instead of only renting them.

How it works, simply: think of the model as a vast library. It does not read every book for every question. Instead it walks straight to the few shelves relevant to the request. That design lets ML4 be enormous yet efficient when answering a single prompt.

What changes now in practice: startups and product teams can start testing ML4 on tasks where Mistral claims strong results, like cybersecurity, finance and manufacturing. When weights arrive, well-resourced labs and companies will be able to run the model on their own infrastructure rather than relying solely on external APIs. A practical reality: training and serving a model like this required thousands of GPUs, so self-hosting will be realistic for larger organizations, not for a laptop or a small cloud bill.

The open question is simple and important: will independent teams turn preview access and the promised weight release into competitive products that truly challenge closed systems? We should see the answer unfold over the coming months.