CERN and Mistral AI Build 'Flare' — the First Foundation Model for Particle Physics
European physics lab CERN and French AI company Mistral AI have partnered to build Flare, a large physics model trained on one exabyte of Large Hadron Collider collision data to reconstruct particle events and accelerate scientific discovery.
CERN and French AI company Mistral AI announced a long-term strategic partnership on 15 October 2026 to develop Flare, billed as the first foundation model for collider physics. The model will apply the same architecture behind large language models — but trained on collision data rather than text — starting with roughly one exabyte of data from CERN's Large Hadron Collider, one of the world's largest scientific datasets. CERN identified four core uses for Flare: reconstructing what physically happened inside a collision, identifying the particles produced, flagging unexpected signals that could hint at new physics, and helping physicists evaluate designs for next-generation detectors. Flare is intended to replace a patchwork of highly specialised hand-coded algorithms currently used for each of those tasks. For Mistral, the partnership provides access to an extraordinary scientific dataset as the French startup seeks to establish itself on the scientific frontier and compete with larger US rivals. CERN's collaboration with a European AI company also reflects a broader push to keep cutting-edge AI capabilities within European institutions. The partnership attracted wide coverage in science and technology media, with analysts noting it could mark a turning point in how big-science experiments handle the enormous data volumes generated by modern particle colliders.
