Mistral Unleashes “Le Chonk,” a 1-Trillion-Parameter Open-Weight AI Model

French AI startup Mistral AI made a major move in the open-source AI race on October 29, 2026, releasing the open weights for its flagship model, Mistral Large 4—affectionately nicknamed “Le Chonk”—on Hugging Face.

The model uses a Mixture-of-Experts (MoE) architecture, with 1.05 trillion total parameters and approximately 52 billion active parameters per token. That makes it one of the largest openly available AI models in the world.

The release includes:

• 1.06 TB of model weights across 54 shards
• A Mistral Ecosystem Open-Weight License, with no access gate
• A focus on agentic coding and cybersecurity

Mistral claims the model delivers state-of-the-art performance at a fraction of the compute cost of closed-source rivals. Following three additional weeks of reinforcement learning beyond its October 6 preview, its Terminal-Bench 4.0 score rose from 28.3% to 41%.

According to Mistral’s VP of Science, Pierre Stock, the open-weight approach allows organizations to deploy the model on-premises. This could give security teams AI-powered defenses that “can’t be cut off by definition.”

The release is widely seen as Europe’s answer to the dominance of large Chinese open-weight AI models.

Source: TechCrunch — Mistral releases weights for its trillion-parameter AI model, targeting cybersecurity