The Pulse
CERN and Mistral Plan a Physics Model for LHC Collisions
CERN and Mistral signed a statement of intent to develop a Large Physics Model trained on billions of simulated Large Hadron Collider collisions. The proposed model would support collision reconstruction, particle identification, anomaly de
AI.info Team · Written with AI ·
Billions of simulated proton-proton collisions will form the training material for a proposed Large Physics Model from CERN and Mistral, the partners announced on October 15. The organizations signed a statement of intent for a long-term partnership to develop foundation models for scientific discovery, beginning with high-energy physics. The announcement sets out an ambition, not a finished model: CERN and Mistral have not announced a release date or technical specifications.
One model for several LHC tasks
CERN says the proposed model would be the first foundation model for collider physics. It is intended to handle several jobs now carried out across different parts of the analysis process: reconstructing particle collisions from detector signals, identifying particles, flagging unexpected signals and assessing future detector designs. The partners say the project will combine CERN’s physics expertise, scientific data and large-scale computing with Mistral’s experience training large models.
Enrica Porcari, CERN’s Chief Information Officer, described the laboratory as a testing ground for AI in physics. “High-energy physics provides a unique environment in which artificial intelligence can be developed and tested with exceptional rigour. We are partnering with Mistral, a leading European AI company, to advance AI capabilities that can serve the whole physics community and be shared,” she said.
Replacing a chain with a shared model
Today, LHC collision reconstruction relies on a sequence of specialised algorithms. Each stage transforms detector signals into particles and other objects that physicists use in their analyses. Machine learning already improves individual tasks in that chain; CERN’s proposal is to train a model that learns across it and can support several tasks in one system.
That design could give researchers a more unified tool for working with collision data, but the announcement does not report test results or claim that the approach will outperform existing methods. The model’s training plan, as described by CERN, starts with simulated collisions. How researchers will test its performance against real detector data, and how they will measure whether it can reliably flag unfamiliar signals, remains unspecified.
The scale comes with a rising workload
CERN says it has stored around one exabyte of experimental data from the LHC—about one million terabytes. The number of collisions is expected to increase tenfold with the High-Luminosity LHC, which CERN expects to begin operating around 2030. That projected increase in collisions makes the ability to process and interpret detector data a central challenge for the laboratory.
Johannes Brandstetter, Mistral’s Vice-President of AI for Engineering, said the partnership aims to explore how large models could address the growing complexity of particle physics. Both organizations present the work as a potential contribution to scientific discovery, not as evidence that the model has found new particles or produced a physics result. CERN also says the effort could eventually inform AI approaches for a wider scientific community, though high-energy physics is the initial focus.
A signed intent, with key terms still open
The announcement does not specify the model’s architecture, compute requirements, budget, development milestones or conditions for access to the eventual system. It also does not say whether the model or its training materials will be released publicly. Porcari said the AI capabilities should “serve the whole physics community and be shared,” but CERN’s announcement does not define what sharing will mean in practice.
Those details will determine how physicists can evaluate the model and whether teams outside the partnership can use it. For now, the concrete commitment is a signed statement of intent, a focus on collider physics and a plan to train on billions of simulated LHC collisions.