French AI laboratory Mistral has announced Mistral Large 4, nicknamed Le Chonk, a one trillion-parameter open-weight model available in preview ahead of a full release scheduled for the end of the month. According to the company, the model was trained from scratch without relying on distillation and is optimized for coding, cyberdefense, manufacturing, electrical engineering, and finance. Mistral presents the release as the most capable open-weight model built outside of China, offering performance close to leading proprietary US systems.

The launch comes alongside growing momentum for the company, which recently raised $3.3 billion at a $24 billion valuation and reported a 20-fold increase in earnings over the past year. Mistral generates revenue by charging cloud usage fees and providing custom engineering services, avoiding the closed-weight subscription model used by American rivals.

Le Chonk arrives amidst tightening US restrictions on AI model exports and heightened international geopolitical tensions. Mistral executives position the open-weight release as a cost-effective alternative for businesses seeking independent AI capabilities that eliminate reliance on proprietary US or Chinese providers.

Why it matters

  • Provides enterprises with a high-parameter, open-weight alternative optimized for domain-specific tasks like coding and industrial engineering.

  • Reduces exposure to US regulatory export controls and vendor lock-in from proprietary cloud model providers.

  • Demonstrates that non-US labs can train competitive trillion-parameter models from scratch with significantly less capital.

Source: wired.com