Mistral launches Large 4 public preview with 1.05T total parameters, 49B active parameters and 1M context, with weights due later this month

What changed

Mistral launched the public preview of Mistral Large 4 on October 6. The company describes it as a multimodal MoE with roughly 1.05 trillion total parameters, 49 billion active parameters and a 1M context window, targeting coding, agents, cybersecurity, finance and multimodal reasoning. The preview is available through Mistral Studio/API, while model weights are planned for release later this month and are not yet generally downloadable. Mistral says the model was trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs in its own European datacenters.