Mistral releases weights for trillion-parameter ML4 model

What happened: Mistral has published the weights of Mistral Large 4, nicknamed Le Chonk. The one-trillion-parameter model was first introduced earlier in October. TechCrunch reported the release on 29 October.

The numbers: Mistral says ML4 performs comparably to GLM-5.3, a Chinese open-weight model, and uses 47% fewer tokens per task than GLM-5.3. It also says ML4 outranks all other open models on Artificial Analysis' Coding Agent Index. These figures are Mistral's own and have not been independently verified.

The details: Weights are available on Hugging Face and through Mistral's API. Organisations can deploy the model on-premises. Mistral says it reaches state-of-the-art performance at a fraction of the compute and capital of closed labs.

Why it matters: Mistral positions ML4 for sovereign cyber defence. It targets enterprises and governments that want to deploy and control the model on their own infrastructure. The company also highlights agentic coding.

In their words: Pierre Stock, Mistral's Vice President of Science, said: "Access that can't be cut off by definition because the model is open-weighted."

The context: Stock also addressed the risks of rogue AI agents. He said Mistral does not want to fearmonger people, and that the problem is solvable.

The catch: TechCrunch calls the model open-weight but names no licence, so the specific licence terms are unconfirmed. Benchmark results beyond Mistral's own comparisons are also unconfirmed.

What's next: Stock said Mistral wants to make this cyber capability available to everyone as soon as possible.