The Flare timeline
Every dated step towards a foundation model for collider physics, with its source, and the dates announced for what comes next.
Gold dots are news about the CERN and Mistral model itself; the rest is the work it builds on. We add a row the day something happens, newest at the bottom, and link the source for each.
So far
-
CERN starts Next Generation Triggers
A five-year project with ATLAS and CMS, paid for by a $48 million grant from the Eric and Wendy Schmidt Fund, to bring AI into the systems that decide which collisions to keep.
Source: CERN Courier
-
OmniJet-α calls itself the first cross-task foundation model for particle physics
A Hamburg group shows one model pretrained to generate jets can be tuned to classify them. Much of the later collider work builds on the same idea.
Source: arXiv 2403.05618
-
A position paper coins "Large Physics Models"
Twenty-two physicists and philosophers propose physics-specific foundation models built by the community. Their sense of the term is broader than the collision model CERN now plans.
Source: arXiv 2501.05382
-
CERN publishes an AI strategy for the whole laboratory
Four goals, one of them enabling AI at scale through partnerships with member states, EU programmes and industry.
Source: CERN
-
CMS shows a fully machine-learned reconstruction
A single model trained on simulated collisions replaces much of the hand-written particle-flow logic and matches the traditional algorithm, with jets 10 to 20% more precise in some ranges.
Source: CERN
-
Mistral agrees to buy Emmi AI
The Austrian physics-AI company, spun out of JKU Linz, brings its co-founder Johannes Brandstetter and a team of more than 30 researchers to Mistral.
Source: Mistral
-
Mistral launches its physics AI work
Mistral names ASML, Airbus, Safran and Siemens Energy as partners for engineering simulation. Its post says physics AI is not an LLM trained on simulation data.
Source: Mistral
-
The LHC stops for its long upgrade
Run 3 physics was planned to end on 29 June 2026, starting Long Shutdown 3, the works that turn the LHC into the High-Luminosity LHC.
Source: CERN
-
A "prompting" model maps any reconstruction step to any other
One tokenised model takes any mix of tracks, calorimeter cells and particles and predicts the others. It is the closest published idea to learning the whole reconstruction chain at once.
Source: arXiv 2609.31862
-
· Flare
CERN and Mistral sign a statement of intent
A long-term partnership whose first aim is a pretrained Large Physics Model for collider physics, trained on billions of simulated proton-proton collisions. No budget, timeline or licence is given.
Source: CERN
-
· Flare
Reuters reports the model as "Flare"
Reuters' headline names it Flare and quotes a CERN spokesperson: the same idea as the large language models behind generative AI, trained on collision data instead of text.
Source: Reuters (via KSL)
Expected
Dates someone has announced that haven't arrived. None of them is a date for Flare: there isn't one yet.
-
· expected
Mistral Large 4's weights are due (not Flare)
Mistral says the weights of its trillion-parameter general model drop at the end of October. Reports give 27 or 31 October. It is a separate model, but often confused with Flare in coverage.
Source: Mistral
-
· expected
The High-Luminosity LHC is due to start
CERN's schedule puts the start in June 2030. CERN says it will bring about ten times the LHC's original design collisions, the data problem a model like Flare is meant to help with.
Source: CERN
Hear when Flare gets a date
Nothing about Flare is public yet beyond the announcement. Leave your email and we'll write once, when the first paper, model or date appears, with what it says. Then we delete your address.