CERN and Mistral announced on October 15, 2026, that they had signed a statement of intent for a long-term partnership to develop AI models for scientific discovery. The partnership’s initial focus is high-energy physics, with a planned model called Flare intended to support analysis of collisions at the Large Hadron Collider (LHC). CERN’s announcement describes a proposal, not an operating model.

What CERN and Mistral plan for Flare

CERN calls Flare a “Large Physics Model”: a single pretrained model intended to learn across parts of the collider-analysis process. Today, CERN says, collision reconstruction relies on chains of specialized algorithms. These turn detector signals into particles and other objects physicists can use in their analyses. Flare is meant to support several tasks across that work rather than one step alone.

CERN’s proposed training input is billions of simulated proton-proton collisions. The intended tasks are:

  • reconstructing collisions;
  • identifying particles;
  • flagging unexpected signals; and
  • evaluating future detector designs.

Because these are proposed uses, they describe the partnership’s aims—not demonstrated results. CERN said the partners want AI capabilities that can serve and be shared with the physics community. The announcement does not specify whether Flare itself will be open-weight or publicly downloadable, or identify a license.

Earlier CERN AI research offers context

CERN’s June 13, 2024, explainer describes machine-learning searches by the ATLAS and CMS collaborations for unusual jet signatures and anomalous collision features. In the CMS results discussed there, the methods had different sensitivities, and no single algorithm performed best across them. CERN’s explainer offers a concrete example of earlier, specialized AI research at the laboratory; those searches were separate from the planned Flare project.

The LHC data scale behind the proposal

CERN reports that about one exabyte of LHC experimental data has been stored to date—equivalent to one million terabytes. The planned model’s training data, by contrast, is described as simulated collisions. CERN also expects the High-Luminosity LHC (HL-LHC) to increase the number of collisions tenfold when it begins operating around 2030.

Reuters reported that the partners had disclosed no project budget or delivery timetable in coverage of the announcement. Reuters’ report, republished by KSL, also covered the partnership.