Mistral Large 4 goes open weight: a 1-trillion-parameter model built for coding

Mistral has taken the wraps off its biggest model. Mistral Large 4, nicknamed "Le Chonk", is now available as open weights, a few weeks after it entered public preview. It's a 1-trillion-parameter model that handles images as well as text, and Mistral says the extra training since the preview has made it the strongest open-weight model for coding.

What changed since the preview

Mistral says it spent the time on more reinforcement learning, mostly aimed at coding and defensive security. The company's own numbers:

• Terminal-Bench 4.0: 41%, up 13 points from the preview, while using about 30% fewer output tokens.
• Artificial Analysis Coding Agent Index: 48%, roughly tied with the top open model while using 47% fewer tokens per task.
• Cyber Index: 51% on finding, reproducing and patching vulnerabilities, with a 96.6% average refusal rate on malicious hacking requests.

There are now three reasoning modes, low, medium and high, so developers can trade depth for speed and cost. In blind human comparisons, graders preferred it over other leading open models, with only Claude Opus 5.5 ranked higher among the models tested.

Price and license

Through the API it costs $0.70 per million input tokens and $2.10 per million output tokens. The weights are on Hugging Face under a modified MIT license: free to use, fine-tune and sell, except that companies offering it as a hosted model service need a separate commercial license.

My take

What interests me here is the token efficiency. A model that solves the same coding task with half the tokens is cheaper to run every single day, which matters more to small teams than one extra benchmark point. Most people won't run a trillion-parameter model at home, but having it open means more apps and studios can host their own version instead of relying on a single provider.

Read Mistral's announcement for the full benchmark charts.