AI Models Powered by Physics Propel Science Forward

Researchers develop foundational AI models that can apply physics knowledge across disciplines

Jan. 27, 2026 at 10:23pm

Got story updates? Submit your updates here. ›

Researchers from the Polymathic AI collaboration have created two new AI models, Walrus and AION-1, that are trained on real scientific datasets to tackle problems in astronomy and fluid-like systems. Unlike most AI models focused on a specific subfield, these 'foundational models' can apply their physics knowledge to a wide range of problems, accelerating scientific discovery and helping researchers work with limited data or resources.

Why it matters

Typical AI models are trained on text or images, but these new physics-based models can cross-apply their learnings to completely different fields, giving researchers a head start when facing new challenges. By training on massive datasets spanning multiple disciplines, these foundational models provide a powerful starting point that can adapt to various scientific scenarios.

The details

The Walrus model is trained on 15 terabytes of data describing fluid dynamics in systems ranging from merging neutron stars to acoustic waves. AION-1 is trained on over 100 terabytes of astronomical data from surveys like the Sloan Digital Sky Survey, allowing it to extract more information from low-resolution galaxy images. These models learn the underlying physics principles, rather than just the specifics of a particular situation, enabling them to be applied broadly.

  • The Polymathic AI team recently announced Walrus in a preprint on arXiv.org and presented AION-1 at the NeurIPS conference in San Diego in 2026.

The players

Polymathic AI

A multidisciplinary collaboration of researchers developing physics-based AI models.

Michael McCabe

A research scientist at Polymathic AI and the lead developer of the Walrus model.

Miles Cranmer

A member of the Polymathic AI team and a researcher in the Department of Applied Mathematics and Theoretical Physics at the University of Cambridge.

Payel Mukhopadhyay

A researcher at the University of Cambridge's Institute of Astronomy and part of the Polymathic AI team.

Liam Parker

The lead researcher for the AION-1 model and a researcher at the University of California, Berkeley.

Got photos? Submit your photos here. ›

What they’re saying

“Maybe you have new physics in your scenario that your field isn't used to handling, and just can't burn the time working through all the possible models that might fit your scenario. Our hope is that training on these broader classes makes something that is both easier to use and has a better chance of generalising for those users, as the 'new' physics to them might be something another field has been handling for a while.”

— Michael McCabe, Research Scientist, Polymathic AI

“I continue to be awed by the fact that a multi-disciplinary physics foundation model works at all, let alone at this level. This question is part of what motivated us to start Polymathic in the first place, and Walrus feels like a nice checkpoint in this direction.”

— Miles Cranmer, Researcher, Department of Applied Mathematics and Theoretical Physics, University of Cambridge

“Walrus feels like a real step toward general-purpose AI for physical simulation-a single foundation model you can adapt across many scientific problems instead of re-training from scratch each time. And because we've open-sourced the code and data, I'm genuinely excited to see what the community builds on top of it.”

— Payel Mukhopadhyay, Researcher, Institute of Astronomy, University of Cambridge

“I think our vision for some of this foundation model is that it enables anyone to start from a really powerful embedding of the data that they're interested in … and still achieve state-of-the-art accuracy without having to build this whole pipeline from scratch.”

— Liam Parker, Researcher, University of California, Berkeley

“We want to bring all this AI intelligence to the scientists who need it.”

— Shirley Ho, Principal Investigator, Polymathic AI

What’s next

The Polymathic AI team plans to continue developing and refining their physics-based foundational AI models to make them more accessible and useful for scientists across various disciplines.

The takeaway

These new physics-powered AI models represent a significant advancement in scientific computing, providing researchers with a powerful tool that can adapt to a wide range of problems and accelerate discovery by leveraging cross-disciplinary knowledge. The open-sourcing of the Walrus model's code and data further enhances the potential impact of this technology on the scientific community.