Deyron Labs graphic: OpenAI says GPT-6 Astra helped build a candidate lattice definition of the Standard Model; guest post by Nathaniel Craig on 15 November; Craig calls it an answer in principle and says expert dialogue is still required

What happened

On 15 November 2026 OpenAI published “A step toward defining the Standard Model on a lattice”, a guest post by Nathaniel Craig, a physicist at UC Santa Barbara who is described as a visiting scientist at OpenAI. It is the first guest post on OpenAI’s new science blog. Craig describes how he worked with GPT-6 Astra on a candidate definition of the Standard Model of particle physics on a lattice, a way of putting a physical theory on a discrete grid. We read the post on 16 November 2026.

Key details

  • The claim: Craig reports “a candidate nonperturbative lattice definition” of the Standard Model, valid under a limit on field strengths at the scale of the grid. He calls the result “far more an answer in principle than an answer in practice”.
  • Role of the model, as described by the author: Craig says Astra “helped turn that strategy into a construction and took the lead in developing the detailed proof”. He also says it suggested how to carry the fields over to finer grids and showed that the comparisons converge.
  • Outside checks, as described in the post: Craig says the proof was examined by outside specialists, naming Golterman, Kronfeld and Shamir, who asked questions. The post does not present this as a formal independent evaluation.
  • The paper: the post links to a preprint from October that says it was prepared with the aid of GPT-6 Astra.
  • Background given in the post: the question of defining the Standard Model nonperturbatively is more than 50 years old, and the problem of chiral fermions on a lattice had stayed open for more than 25 years, according to the post.
  • Open points: Craig considers some questions still open, including whether the definition respects reflection positivity at finite spacing, and writes that “scientific dialogue among experts is required”.
  • The new blog: a separate OpenAI post, “Our new blog for science”, launches the blog for researchers. It recounts earlier results that OpenAI itself reports, which this article does not independently verify.
  • Not stated: any product availability, pricing, or independent peer review of the result.

Why it matters

The post is an example of a lab presenting a language model as a working partner in theoretical physics, with a named scientist describing the model’s contributions in detail. It is also an account written by the author and published by the model’s maker, so the central claim, that a candidate construction exists and its proof holds up, rests on the author’s description and on the scrutiny of experts that has yet to be reported.

For readers following AI in research, the useful signals are the author’s own caveats: this is a candidate definition, not an accepted result, and the community’s evaluation is the next step.

Disclosure: Claude, made by Anthropic, was one of the AI tools used to research and draft this article. OpenAI is a competitor of Anthropic.

Sources

  1. A step toward defining the Standard Model on a lattice Primary source , OpenAI (guest post by Nathaniel Craig), 15 November 2026
  2. Our new blog for science OpenAI, 15 November 2026