The company says the preview API is available now through Mistral Studio and that model weights are expected by the end of October. Mistral says it trained the model on 3,800 NVIDIA Grace Blackwell GPUs in its own European data centers. Its announcement presents results in coding, agentic workflows, cybersecurity, multimodal understanding, and enterprise tasks; these are company-reported claims, with some coding benchmark figures evaluated privately ahead of public harness availability. The release creates a concrete test of whether an open-weight model can pair frontier-scale capability with customer control and self-deployment. https://mistral.ai/news/mistral-large-4
Raw SignalOct 6, 20261 min read1 source
Introducing Mistral Large 4
In one minute
Mistral has announced a public preview of Mistral Large 4, a natively multimodal model with 1 trillion parameters and 49 billion active parameters.