Flare papers and code
No paper on Flare has been published yet. These are the papers it builds on, oldest first, each with a line on why it matters.
Papers on Flare itself
None yet. CERN and Mistral have published a statement of intent (CERN's release) but no paper, code or model card. We check arXiv, CERN's news and Mistral's news page, and add the first one here the day it appears.
Learning to reconstruct whole collisions
CERN's release says a foundation model could learn across the whole reconstruction chain. These are the steps towards that so far.
-
Particle-flow reconstruction and global event description with the CMS detector
The hand-built particle-flow algorithm a learned model would have to match or beat.
-
MLPF: Efficient machine-learned particle-flow reconstruction using graph neural networks
The first version of the machine-learned particle flow CMS showed working in 2026.
-
Machine-learned particle flow as a foundation model for collider physics
Argues the reconstruction model's inner representation already serves other tasks, the definition of a foundation model.
-
Prompting Particle Physics: Tokenized Multi-modal Foundation Models for Combinatorially Many Tasks
One model for any step from detector hits to particles, the nearest published idea to Flare.
Pretraining on collision data
The jet and event foundation models that came before Flare, oldest first.
-
Particle Transformer for Jet Tagging
Introduced JetClass, the 100-million-jet dataset most later models pretrain on.
-
Masked Particle Modeling on Sets: Towards Self-Supervised High Energy Physics Foundation Models
Brought masked-token pretraining to sets of particles.
-
OmniJet-α: The first cross-task foundation model for particle physics
The first model shown to transfer from generating jets to tagging them.
-
Solving Key Challenges in Collider Physics with Foundation Models
OmniLearn, one pretrained model tuned for many collider tasks.
-
Aspen Open Jets: Unlocking LHC Data for Foundation Models in Particle Physics
178 million real CMS jets for pretraining, an answer to the worry about learning only from simulation.
-
HEP-JEPA: A foundation model for collider physics using joint embedding predictive architecture
A different pretraining recipe, predicting in an abstract space.
-
OmniLearned: A Foundation Model Framework for All Tasks Involving Jet Physics
Trained on more than a billion jets, with checkpoints released.
-
EveNet: A Foundation Model for Particle Collision Data Analysis
Pretrained on whole events, 500 million of them, rather than single jets.
-
Predict before you train: Scaling Laws for particle physics foundation models
Evidence that bigger particle-physics models improve predictably, the bet behind building Flare large.
-
A Lightweight Foundation Model for Collider Physics with Multi-Domain Adaptation
NEXUS, a 3-million-parameter model for real-time use.
Finding the unexpected
Flagging unexpected signals is one of the four jobs CERN lists for the model.
-
Searching for Anomalies with Foundation Models
How a pretrained model can be turned into a search for physics nobody predicted.
Overviews and arguments
Where to start reading.
-
The paper that put "Large Physics Models" into print, in a broader sense than CERN's.
-
Foundation models for high-energy physics
The first review of the whole topic.
-
Machine learning for the LHC physics program: a 2025-2026 stocktake
A survey of where machine learning at the LHC stands. Its author says much of the text was written with an AI system.
Titles are the authors' own, as on arXiv. Missing something? Email [email protected].
Hear the day the first paper lands
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.