OpenAI's GPT-6 Astra helps build proof that could solve a 50-year-old physics problem

OpenAI's GPT-6 Astra helps build proof that could solve a 50-year-old physics problem

OpenAI's GPT-6 Astra helped produce a proof that could solve a 50-year-old problem in particle physics. Physicist Nathaniel Craig of UC Santa Barbara, currently a visiting scientist at OpenAI, presented the approach on the OpenAI blog and as a preprint paper. Craig leads the particle theory group at UCSB and is a permanent member of its Kavli Institute for Theoretical Physics.

The background: physicists can simulate parts of the Standard Model, the theory describing all known particles and fundamental forces, on a computer grid called a lattice. That has already led to breakthroughs such as calculating the proton's mass from first principles. But putting the full theory on a lattice has been impossible. The weak force, which governs certain types of radioactive decay, treats left- and right-handed particles differently. On simple lattices, this asymmetry creates unwanted ghost particles. Special lattice particles called overlap fermions can filter them out, but for over 25 years nobody could set their phase correctly.

Craig had a strategy to fix the phase by comparing progressively finer lattices, but he could not prove it worked in all the cases it needed to cover. GPT-6 Astra turned the idea into a working construction and took the lead on the detailed proof.

Lattice field theorists Maarten Golterman, Andreas Kronfeld and Yigal Shamir reviewed the proof and raised questions that Craig says helped build confidence in the result. He calls the work "much more a beginning than an end" and says a lot remains before realistic simulations are possible.

The Decoder places the result in a run of AI advances in mathematics that started in January 2026. In May, OpenAI's disproof of Erdős's unit distance conjecture became the first AI proof worthy of math's top journal. Days later, Google Deepmind's AlphaProof Nexus solved nine of 353 open Erdős problems for a few hundred dollars, and by October its results had passed peer review. The most far-reaching claim came in September, when OpenAI submitted a 166-page proof for the Navier-Stokes Millennium Prize Problem and the Clay Mathematics Institute said it had "apparently been settled". That solution remains under review, and mathematicians have questioned whether OpenAI fully understands its own proof. Across the field, mathematicians have mixed feelings as AI keeps cracking unsolved problems.

Key facts

  • GPT-6 Astra helped produce a proof that could solve a 50-year-old particle physics problem: putting the full Standard Model on a lattice.
  • The sticking point was the phase of overlap fermions, which nobody could set correctly for over 25 years; overlap fermions filter out the ghost particles that the weak force's left/right asymmetry creates on simple lattices.
  • Nathaniel Craig of UC Santa Barbara, a visiting scientist at OpenAI, had a strategy based on comparing progressively finer lattices; GPT-6 Astra turned it into a working construction and took the lead on the detailed proof.
  • Lattice field theorists Maarten Golterman, Andreas Kronfeld and Yigal Shamir reviewed the proof and raised questions; Craig says a lot remains before realistic simulations are possible.

Why it matters

Lattice simulations have already produced results like a first-principles calculation of the proton's mass, but the full Standard Model has stayed out of reach because the weak force treats left- and right-handed particles differently. If the proof holds, it could remove a roadblock that has stood for 50 years. It also adds to a run of AI mathematics results that began in January 2026, from OpenAI's May disproof of Erdős's unit distance conjecture to Google Deepmind's AlphaProof Nexus, which solved nine of 353 open Erdős problems.

Who it affects

Mainly lattice field theorists and particle physicists who want to simulate the full Standard Model, including the weak force. It also bears on the wider debate among mathematicians, who have mixed feelings as AI keeps cracking unsolved problems.

How to use it

There is nothing to run or install here. This is a research result, presented by Craig on the OpenAI blog and as a preprint paper. Physicists who work on lattice chiral fermions are the people who would build on the construction.

How solid is it

The claim is hedged: the proof "could solve" the problem, and the source does not say it is solved. Three lattice field theorists (Golterman, Kronfeld and Shamir) reviewed it and raised questions that Craig says built his confidence. Craig, who is a visiting scientist at OpenAI, is also the one presenting the work. The report comes from The Decoder, a secondary outlet, which points to the OpenAI blog and preprint as the primary sources.

Risks and caveats

Craig himself calls the work "much more a beginning than an end" and says a lot remains before realistic simulations are possible. The wider context urges caution: OpenAI's 166-page Navier-Stokes proof, submitted in September, remains under review, and mathematicians have questioned whether OpenAI fully understands its own proof.

“much more a beginning than an end”

— Nathaniel Craig, physicist at UC Santa Barbara and visiting scientist at OpenAI