Mistral has opened a public API preview of Mistral Large 4 (ML4) and says it plans to release the model weights by the end of October. The company describes the model as having one trillion parameters, with 49 billion active, and says the preview runs in its European data centers.
Mistral Large 4 capabilities and evaluations
Mistral positions ML4 as an open-weight model for coding, agent workflows, and multimodal understanding. It reports benchmark results across cybersecurity, software engineering, visual grounding, and professional work. For example, the company says the model solved 93% of Cybench challenges and scored 82% on a test involving reproducing and patching a real software vulnerability. These are reported evaluation results, not a guarantee of performance in every deployment.
Open weights: more control, questions remain
Once the weights are released, organizations are expected to be able to deploy the model in private-cloud or on-premise environments and apply their own policies. That control may matter to teams conducting security research or needing auditable systems. Mistral also says training data included more than 160 languages and that deployments will be available in multiple regions.
The weights are not yet available. Mistral says it is conducting red-team testing with cybersecurity partners and plans to share more about the architecture, benchmarks, and post-training methods. Open questions include whether independent evaluations will reproduce the reported results, and how safety controls, deployment costs, and long-term availability will work in practice.