CERN and Mistral AI plan foundation model for collider physics

CERN and Mistral AI have signed a statement of intent for a long-term partnership centred on developing a Large Physics Model to analyse particle collisions.

CERN announced the agreement on October 15. The laboratory will contribute expertise in fundamental physics, scientific data and large-scale computing, while Mistral AI will provide large-model training capabilities.

Large Hadron Collider detector data currently passes through a sequence of specialised algorithms. These progressively reconstruct particles and other objects for physicists to analyse. Machine learning already supports individual stages. The partners aim to bring several tasks together within a single pretrained model.

Alongside reconstructing collisions, the Large Physics Model is intended to identify particles, flag unexpected signals and help evaluate future detector designs. Work will initially focus on high-energy physics, with possible applications in other scientific fields considered a longer-term direction.

CERN's information technology director, Enrica Porcari, linked the partnership to developing tools for the wider physics community. Mistral's vice president for engineering AI, Johannes Brandstetter, highlighted the opportunity to address the increasing complexity of particle physics.

CERN says it already stores around one exabyte of experimental collider data, equivalent to one million terabytes. Following the High-Luminosity LHC upgrade, expected to begin operating around 2030, the number of collisions is projected to increase tenfold.

The LHC occupies a twenty-seven-kilometre ring tunnel. Separately, CERN is considering the Future Circular Collider, with a circumference of about ninety-one kilometres. A 2025 feasibility report estimated the first stage at fifteen billion Swiss francs, beginning with an electron-positron machine and envisaging a later proton-proton collider at about one hundred teraelectronvolts.