CERN and Mistral sign agreement to build AI for particle-physics research

Particle-collision analysis could gain a single AI model supporting several jobs now handled through specialized algorithms. CERN and Mistral signed a statement of intent on October 15, 2026, to establish a long-term scientific AI partnership. Their initial project is a Large Physics Model trained on billions of simulated collisions at the Large Hadron Collider.

The agreement combines CERN’s expertise in fundamental physics, scientific data and large-scale computing with Mistral’s large-model training capabilities. The partners intend to develop foundation models for scientific discovery: pretrained systems designed to support multiple tasks. High-energy physics is the starting point, with wider scientific applications a longer-term possibility.

CERN describes the proposed system as the first foundation model for collider physics. That remains a development ambition: the announcement does not release a model or demonstrate an improvement over existing algorithms.

Learning across the collision-analysis chain

Today, reconstructing an LHC collision involves a chain of specialized algorithms. These progressively turn detector signals into particles and other objects researchers use in physics analysis. Machine learning already improves performance on individual tasks; the partnership builds on that work rather than introducing AI to the process.

Instead of focusing only on a specialized step, a foundation model could learn across the whole reconstruction chain and support multiple tasks. CERN presents that as a potential advantage. Training would use billions of simulated proton-proton collisions, with intended applications spanning analysis and future equipment:

• Reconstruct particle collisions and identify particles.
• Flag unexpected signals.
• Evaluate future detector designs.

High-energy physics provides a unique environment in which artificial intelligence can be developed and tested with exceptional rigour.

Enrica Porcari, CERN’s Chief Information Officer

A larger collision workload is coming

CERN says it has stored around one exabyte of experimental LHC data to date—one million terabytes. That archive is distinct from the simulated collisions planned for training. It illustrates the scale of scientific data the laboratory already handles.

The workload is expected to grow substantially. CERN forecasts a tenfold increase in collisions with the HL-LHC, which is expected to begin operating around 2030. The laboratory identifies this combination of scale and scientific precision as a demanding setting for developing and testing AI.