CERN and Mistral sign letter of intent for a "Large Physics Model" for the LHC
CERN and the French AI company Mistral have signed a letter of intent to develop a "Large Physics Model" – according to the announcement, the first foundation model for collision physics, trained on billions of simulated…
CERN and the French AI company Mistral have signed a letter of intent to develop a "Large Physics Model" – according to the announcement, the first foundation model for collision physics, trained on billions of simulated proton-proton collisions at the LHC. The letter of intent is nonetheless not a contract: no timelines, amounts, or milestones have been published, and the model must still prove that it holds up against real detector data.
CERN, the European laboratory for particle physics, and Mistral, a European developer of large-scale AI models, announced on 15 October 2026 that they have signed a letter of intent for a long-term strategic partnership focused on foundation models for scientific discovery (Mirage News). Reuters provided its confirmation of the announcement the same day (MSN/Reuters).
The practical substance is an attempt to replace a chain of specialized algorithms with a single pre-trained model. According to the joint announcement, the model will be trained on billions of simulated proton-proton collisions at the Large Hadron Collider and support multiple tasks at once: reconstruction of particle collisions, particle identification, flagging of unexpected signals, and evaluation of future detector designs (Mirage News).
Why this is relevant right now
CERN states that the laboratory stores around one exabyte – one million terabytes – of experiment data from the LHC, one of the largest scientific data volumes in the world. In addition, the number of collisions is expected to increase tenfold with the High-Luminosity LHC (HL-LHC), which is expected to be in operation around 2030 (Mirage News). The coming data-volume leap is the concrete background for rethinking the reconstruction pipeline.
From an algorithm chain to a single model
Today, reconstruction happens in several steps: a chain of specialized algorithms gradually converts detector signals into particles and other objects used in physics analysis. Machine learning already plays an important role in a number of such individual tasks. The point of a foundation model is to build on these advances by learning across the entire reconstruction chain and supporting multiple tasks within one and the same model (Mirage News).
The announcement also emphasizes that the partnership combines CERN's expertise in fundamental physics, scientific data, and large-scale computing with Mistral's competence in large-scale model training (Mirage News).
Mistral's motive
For Mistral, the partnership is also about positioning. Reuters writes that the letter of intent gives the French company access to one of the world's largest scientific datasets, at a company seeking to compete in frontier science (MSN/Reuters).
What the parties say
Enrica Porcari, CIO at CERN, stressed that high-energy physics is a distinctive testing ground for AI: "High-energy physics provides a unique environment where artificial intelligence can be developed and tested with exceptional rigor. We are partnering with Mistral, a leading European AI company, to advance AI capabilities that can serve the entire physics community and be shared," she said in the announcement (Mirage News).
Johannes Brandstetter, Vice President of AI for Engineering at Mistral, emphasized the opportunity: "We are proud to collaborate with CERN to explore what large-scale AI models can contribute to the growing complexity of particle physics. Together, we have the opportunity to push the boundaries of high-energy physics by developing AI capabilities that have the potential to lead to new scientific discoveries," he said (Mirage News).
What remains unclear
Several things are still unresolved. The letter of intent is not a binding contract, and neither party has published timelines, funding amounts, or concrete milestones beyond the agreement itself; according to Cryptobriefing, the parties have had contacts and presentations dating back to 2025 (Cryptobriefing).
The naming question is also open. Although one news headline refers to the model as "Flare," this name appears only in headline slugs and is not confirmed by the announcement texts themselves, which consistently use "Large Physics Model." The name must therefore be regarded as unconfirmed for now.
The most important scientific uncertainty is that the model will be trained on simulated collisions, not real data. As Cryptobriefing points out, models trained on simulated data must prove that they hold up against real detector output (Cryptobriefing). It is precisely this leap from simulation to real detector data that will determine whether the concept works in practice – and it remains unanswered for now.
It should also be noted that all available sources are secondary accounts of the same joint announcement, rather than independent reporting with new details.