CERN and Mistral sign a statement of intent to establish a strategic partnership

The planned model would be trained on billions of simulated proton-proton collisions at the Large Hadron Collider. It is meant to take over reconstruction and particle identification work that some specialised algorithms now handle. It could also flag unexpected signals and help evaluate future detector designs. CERN notes that LHC experimental data stored to date total about one exabyte, and collisions are expected to increase tenfold with the HL-LHC around 2030. Because the document is a statement of intent, it does not announce a finished scientific instrument. Press coverage has labelled the model Flare, although CERN's announcement does not use that name. The effort sits inside CERN's broader planning. The Enlarged Directorate approved a CERN-wide AI strategy in early November 2025, with enabling AI at scale through strategic partnerships among its four goals. Other groups have already built physics models, such as OmniLearned, which trains on more than one billion jets, and EveNet, which was pretrained on 500 million simulated events. Those models mostly start from jets or from particles that standard software has already reconstructed. The CERN plan differs in aiming to learn from raw detector signals across the whole reconstruction chain. For founders, the split of roles is the useful part. CERN supplies physics expertise, scientific data and large-scale computing, while Mistral supplies large-scale model training. Public labs with large proprietary datasets may prefer domain-specific foundation models over generic ones, and a partnership structure like this one gives a company a path into that market. The open questions are how shared access will work for the wider physics community and how the terms will be funded.