CERN and Mistral sign statement of intent to develop Large Physics Model

CERN and Mistral signed a statement of intent on 15 October 2026 for a long-term partnership to develop AI foundation models for physics at CERN's Large Hadron Collider. Their initial plan is a Large Physics Model that could help researchers reconstruct collisions and identify particles as the volume of collider data grows.

How the Large Physics Model would be used

The partners intend to pretrain the model on billions of simulated proton-proton collisions. They say it could reconstruct collision events and identify the particles produced. The public release also names two possible uses: flagging unexpected signals and evaluating designs for future detectors. These are proposed applications of a model the partners plan to develop, rather than reported findings from one already in use.

Reuters, in a report republished by Channel NewsAsia, says the proposed model is intended to replace specialised algorithms used for reconstruction and particle identification. CERN's public release describes today's process as a chain of algorithms that converts detector signals into particles and other objects used in physics analysis. Machine learning already helps with specialised tasks in that chain; the new proposal would apply a pretrained model across a broader set of physics tasks.

Why CERN is planning for more collider data

The scale of the data helps explain the interest. CERN says it has stored around one exabyte of experimental Large Hadron Collider data. It expects the High-Luminosity LHC to begin operating around 2030, with the number of collisions increasing tenfold. That would put more pressure on the methods used to turn detector readings into data physicists can analyse.

CERN's chief information officer described the aim as developing AI capabilities that can serve the physics community and be shared. A Mistral executive described the partnership as a chance to explore large-scale models for particle physics and possible discoveries. Those statements describe the partners' ambitions, not results from the proposed model.

What the agreement commits to

The announcement establishes a statement of intent, not a completed model or a delivery schedule. Reuters reports that the partners disclosed neither a timeline nor a budget; Crypto Briefing likewise reports no funding figure or specific milestones. Crypto Briefing says a model trained on simulations would still need to show that it works with real detector output. The available reports do not describe such a validation result.