Mistral gets the keys to CERN's exabyte: Europe's AI champion teams up on Flare, the first foundation model for collider physics

What's the deal? CERNDealroom has a profile for this one. Try Dealroom → and Mistral AI have signed a statement of intent for a long-term strategic partnership to build "Flare," which CERN says would be the first foundation model for collider physics. The two organisations plan to train a single "Large Physics Model" on billions of simulated proton-proton collisions from the Large Hadron Collider (LHC), using the same underlying approach as large language models but trained on collision data rather than text.

What's the endgame? Flare is meant to replace some of the specialised algorithms CERN currently uses to reconstruct particle tracks and identify particle types, flag unexpected signals in collision data, and help evaluate designs for future detectors. CERN has used AI to turn raw experimental data into analysable events for a decade, but currently relies on multiple separate tools for each step.

Why now? CERN holds about one exabyte — roughly one million terabytes — of experimental data from the LHC, and expects collision volumes to grow tenfold once its ongoing upgrade completes around 2030. Mistral chief executive Arthur Mensch wrote in Le Monde the same day that frontier AI models are "now indispensable to scientific progress," criticising labs that keep their capabilities to themselves. CERN chief information officer Enrica Porcari said the partnership would "advance AI capabilities that can serve the whole physics community and be shared," calling high-energy physics "a unique environment in which artificial intelligence can be developed and tested with exceptional rigour."

What's not disclosed: Neither side has set a budget or timeline for Flare. Both say the resulting models could eventually serve other scientific fields beyond particle physics.

The signal: For Mistral — widely seen as Europe's best shot at catching up with US and Chinese AI labs, and which plans to release a new open-weight model to the public on 27 October — a tie-up with one of the world's largest scientific datasets is a high-profile way to prove out frontier capabilities beyond chatbots and enterprise deployments, while CERN gets a shot at AI tools purpose-built for physics rather than adapted from general-purpose models.

Read more: CERN · Reuters via CNA

Image credit: Chris Mitchell, CC BY-SA 4.0, via Wikimedia Commons