Mistral releases latest AI model for download, giving companies more control over their cybersecurity

Mistral, a French AI developer, has unveiled its latest large-scale AI model, Mistral Large 4 (ML4), known as "Le Chonk". The company claims ML4 offers superior performance in coding and cybersecurity, aiming to provide European organizations with "frontier AI cyber defence capabilities they can fully own and control". ML4 can detect, analyze, and respond to cyber threats in code and IT infrastructure.

Following a preview in early October, Mistral spent three weeks refining ML4, incorporating customer feedback, enhancing efficiency, and customizing the model. Vice President of Science Pierre Stock stated that the open-weight version continues reinforcement learning training and demonstrates substantial gains without reaching a saturation point.

ML4 boasts 1.05 trillion parameters, with 52 billion being activated during tasks, making it more efficient than other AI models. It operates with 47% fewer tokens per task on average compared to the Chinese GLM-5.3 model. ML4's performance on Terminal-Bench 4.0, a test for AI agents' coding abilities in a terminal, increased from 28% to 41% following further training.

The release of ML4 comes at a time when Europe faces increasing cybersecurity threats, such as a recent PHP programming security flaw discovered using AI and cyber-attacks on European governments. By providing the model's weights, organizations can download and run ML4 on their own infrastructure, maintaining greater control over where it runs and how it integrates with their systems, while avoiding potential access restriction risks posed by third-party providers.