Mistral launches 1 trillion-parameter AI model for cybersecurity

French AI lab Mistral unveiled Mistral Large 4 (Le Chonk), a 1-trillion-parameter open-weight AI model focused on cybersecurity. Trained on 4,000 NVIDIA Grace Blackwell GPUs over two months, it's in public preview with weights releasing October 27. Deployed in European data centers, it competes with GPT-6 Astra and supports EU data compliance.