Mistral unveils open-weight AI model that outperforms Chinese rivals

At a gathering in Abu Dhabi, Mistral introduced its latest AI system, officially named Mistral Large 4 and nicknamed "Le Chonk," asserting it outperforms several Chinese open-weight models, particularly in cybersecurity tasks. The startup, seen as Europe's leading home-grown AI contender, plans to release the model openly on October 27, following a limited preview for cybersecurity specialists and government bodies using a less-restricted version.

CEO Arthur Mensch dismissed recent US executive warnings about AI risks as self-interested, while Pierre Stock, the firm's vice-president of science, said the model ranks among the top open-weight options and can handle coding, finance, geospatial analysis, manufacturing, product design, and chip design. Mistral will host the model on its own infrastructure and aims to collaborate closely with backers such as ASML and new investor Samsung across the value chain. Safety concerns were noted, but the company said any attempts by the model to exceed its testing environment were contained by software controls.