OpenAI's GPT-6 Astra tackles the Standard Model's toughest math problem on a lattice
OpenAI’s GPT-6 Astra helped turn a lattice strategy from UC Santa Barbara physicist Nathaniel Craig into a working construction and detailed proof aimed at a barrier that has kept the full Standard Model off a computer grid. Craig, who leads UCSB’s particle theory group, is a permanent member of the Kavli Institute for Theoretical Physics and a visiting scientist at OpenAI; he presented the work on the company’s blog and as a preprint.
Parts of the Standard Model can already be simulated on a lattice, including first-principles calculations of the proton’s mass. The full theory has stayed out of reach because the weak force treats left- and right-handed particles differently, and that asymmetry produces unwanted ghost particles on simple lattices. Overlap fermions can filter those ghosts, but for more than 25 years no one could set their phase correctly. Craig’s idea was to fix the phase by comparing progressively finer lattices; he could not prove it in every case it had to cover. Astra led the proof. Lattice theorists Maarten Golterman, Andreas Kronfeld, and Yigal Shamir reviewed it and raised questions that Craig says strengthened confidence in the result. He calls the work much more a beginning than an end and says a great deal remains before realistic simulations.
The result sits in a 2026 run of AI mathematics that began in January. In May, OpenAI’s disproof of Erdős’s unit distance conjecture was billed as the first AI proof fit for mathematics’ top journal; days later Google DeepMind’s AlphaProof Nexus solved nine of 353 open Erdős problems for a few hundred dollars, with results that had passed peer review by October. In September OpenAI submitted a 166-page Navier–Stokes Millennium Prize proof; the Clay Mathematics Institute said the problem had “apparently been settled,” though the solution is still under review and mathematicians have asked whether OpenAI fully understands its own argument.
Continue in the app — vote & join in ➔