Mistral launches Large 4, a 1T multimodal model with limited access

Mistral is keeping the model behind a guarded endpoint for now, with weights due after safety testing in about three weeks.

Mistral AI has launched Mistral Large 4, a one-trillion-parameter multimodal model that it says is meant to compete with both closed U.S. systems and open models from China. The model is available now only through a guarded public endpoint, and Mistral says it plans to release the weights in about three weeks after safety testing. The company says the model was trained on its own infrastructure using 4,000 NVIDIA GPUs, far fewer than it says Chinese rivals and some closed-source competitors use. Mistral is pitching the system for enterprise uses such as cybersecurity, finance, and chip design, while still arguing that the release keeps it in the frontier-lab category.