Image: OpenAI
OpenAI has published the first guest post on its new science blog, written by physicist Nathaniel Craig, who describes working with GPT-6 Astra on a candidate definition of the Standard Model on a lattice. Astra is the model line we covered when GPT-6.2 Astra and Sol appeared in Google Cloud listings. The underlying preprint has been on arXiv since October 6 and lists Craig as its only author.
Physicist Nathaniel Craig shares how he worked with GPT-6 Astra to develop a candidate lattice definition of the Standard Model. It’s the first post on our new science blog: openai.com/index/step-toward-defining-the-standard-model/
— OpenAI (@OpenAI) November 16, 2026
The problem is older than the model. Physicists usually calculate with the Standard Model by treating its interactions as small corrections, which works to extraordinary precision for processes like the electron's magnetic moment. It fails where the interactions cannot be treated as small, such as the proton's mass, which comes mostly from the strong force inside it. Lattice field theory handles that by placing spacetime on a grid and computing directly, and it has given physicists the proton mass from first principles.
The full Standard Model resists the same treatment because it is chiral: the weak force treats left-handed and right-handed particles differently. Put a left-handed fermion on a lattice in the obvious way and a right-handed partner appears that nobody asked for. Overlap fermions and a rule from Martin Lüscher solved pieces of this for simpler theories, but the phase rule for the full Standard Model stayed open for more than 25 years, according to the post.
Craig writes that he had a strategy but could not prove it covered every case. Astra, he says, suggested carrying a given set of force fields onto successively finer grids, showed the comparisons would converge to a definite answer, and took the lead on the detailed proof. Comparing chiral projectors at successive lattice spacings yields a current that falls off exponentially, and that feeds Lüscher's reconstruction into a gauge-invariant fermion measure on admissible fields, under a bound on field strengths at the scale of the grid.
The arXiv listing backs up the scale of the work: 60 and 63 pages in two parts, 12 figures, and three supplementary documents, including a mathematical companion and a Lean formalization. Craig writes that lattice theorists Maarten Golterman, Andreas Kronfeld and Yigal Shamir read the proof and questioned it before he trusted the result.
What he does not claim is a finished theory. The construction is a candidate, and he says scientific discussion is needed to decide whether it has every property physicists want. Reflection positivity at finite lattice spacing is still open, which matters for connecting the model to a quantum theory where probabilities are conserved. The continuum limit is open too, and OpenAI's earlier proof of Yang-Mills existence is cited as the missing piece. He calls the result an answer in principle rather than in practice, not yet useful for realistic simulations.
The paper credits Craig alone; the model's role comes from OpenAI's account of the work, and the three reviewers are named there rather than in the paper. Nothing has been peer reviewed.
OpenAI started the science blog to describe how researchers use its models, and this post is the first guest entry. The next test is whether lattice physicists accept the measure, and whether the reflection positivity question gets an answer.
Comments