Mistral releases weights for one-trillion-parameter Mistral Large 4

Mistral published the weights of Mistral Large 4 (ML4), a model with one trillion parameters, on 29 October 2026, according to TechCrunch. Mistral said the model is aimed at sovereign cyber defence and agentic coding, for enterprises and governments that want to deploy and control it on their own infrastructure.

The weights are available on Hugging Face and through Mistral's API. Organisations can also deploy the model on-premises.

Mistral said ML4 performs comparably to GLM-5.3, a Chinese open-weight model, while using 47% fewer tokens per task. It also said ML4 outranks all other open models on Artificial Analysis' Coding Agent Index for agentic coding. These figures are Mistral's own and have not been independently verified.

Mistral said the model reaches state-of-the-art performance at a fraction of the compute and capital of closed labs.

Pierre Stock, Mistral's vice president of science, said access to the model cannot be cut off by definition because it is open-weight. Stock said Mistral wants to make this cyber capability available to everyone as soon as possible.

On the risks of rogue AI agents, Stock said Mistral does not want to frighten people and that the problem is solvable.

Mistral first introduced ML4 earlier in October 2026.

TechCrunch describes ML4 as open-weight but names no licence, so its licence terms are unconfirmed. No timeline for future releases has been confirmed.