CERN and Mistral Plan a Particle-Physics Foundation Model Trained on Simulated LHC Collisions

CERN, the particle-physics laboratory on the edge of Geneva, and French AI company Mistral have signed a letter of intent for a long-term collaboration to build a large AI model for particle physics, Swiss IT news site Inside IT reported on October 19, citing a CERN announcement. The first target is a foundation model trained on billions of simulated proton collisions at the Large Hadron Collider.

According to CERN, the model should handle several jobs in analysing collisions: reconstructing events, identifying particles and searching for unexpected signals. It is also meant to help evaluate future detector designs. Today the LHC data are processed by a series of specialised algorithms; the partners want to see whether one large pre-trained model can combine several of those tasks and how such models behave in demanding scientific analysis.

The data volumes are large. CERN says it has stored about one exabyte of LHC experimental data so far, roughly one million terabytes, and the High-Luminosity LHC, expected to start operating around 2030, is set to raise the number of collisions tenfold. CERN chief information officer Enrica Porcari said high-energy physics offers a unique setting to develop and test AI under especially strict conditions, and that the goal is capabilities that can serve, and be shared with, the wider physics community. Johannes Brandstetter, Mistral’s vice-president of AI for engineering, said the partners could push the limits of high-energy physics and build AI that may lead to new discoveries.

What it means in practice

For Switzerland, the deal puts Geneva’s best-known research site at the centre of a scientific-AI effort just as domestic teams push their own open models, as in ETH and EPFL training Apertus 2.0 on Alps. It also echoes Swiss supercomputing teams training AI on NASA data, where domain science rather than chat drives the model design.

A letter of intent is not a delivery schedule, and the announcement did not say where the model will be trained or how openly its weights will be released. Those details will matter to Swiss universities and research computing centres that already run large physics and climate workloads and could become partners, users or rivals for scarce compute.

What to watch: the formal agreement that follows the letter of intent, whether CERN commits to sharing the model openly with institutes in its member states, and how the first results compare with today’s specialised reconstruction algorithms before the High-Luminosity LHC comes online.