On October 15, CERN and Mistral signed a statement of intent for a long-term partnership. The goal is Flare, which they describe as the first foundation model for collider physics.
A language model learns patterns in text. Flare will instead be trained on data from simulated collisions at the Large Hadron Collider. The idea is for one shared model to help scientists reconstruct and identify particles — tasks that currently require several specialised algorithms.
The need grows with the amount of data. CERN already has around one exabyte, or one million terabytes, of experimental data. After a planned upgrade around 2030, the number of collisions is expected to increase roughly tenfold.
But Flare is not available to download or try. The partners have not given a timeline, budget or finished model, and possible use in other sciences remains a goal. Today's announcement is the start of a build, not a completed breakthrough. The interesting part is the direction: the idea of a broad foundation model is now being tested on physics collision data instead of words.