Mistral Large 4 pushes scale—and makes a promise about open weights.

Mistral launched a public preview of Mistral Large 4 on October 6. It describes a natively multimodal mixture-of-experts model with one trillion parameters in total and 52 billion active per token. The company says its model performs strongly across coding, agentic work, visual understanding and specialist enterprise tasks.

The preview is accessible through Mistral's hosted offering. Mistral says it intends to publish model weights by the end of October, following additional red-teaming. As of Cortist's October 19 editorial cutoff, the future release should not be reported as if it had already happened.

A European company offering a model of this scale expands options for enterprises concerned about provider dependency, deployment location and control. Open weights, once actually released under usable terms, can enable independent testing, private infrastructure and customization.

But 'open weights' is not synonymous with completely open development or unrestricted use. Licenses, inference cost, hardware requirements and practical reproducibility shape how much choice customers gain. A trillion total parameters is also not an application outcome.

Cortist sees this as a sovereignty and optionality story. A second credible route to powerful models is valuable even for organizations that never self-host; competition can change bargaining power and product design.

The key editorial discipline is to distinguish current availability from announced intention. Test what is accessible now, then revisit the claim when the weights, license and deployment instructions are actually published.