OpenAI announces three big science discoveries and a new way to work

OpenAI posted three results it calls major discoveries in mathematics, computer science, and physics, less than a month after hundreds of manuscripts it said contained solutions to long-standing mathematics questions. The Verge's report by senior editor Richard Lawler, who joined The Verge in 2021 after several years covering news at Engadget, ran at 11:34 PM UTC on Oct 26, 2026. OpenAI says shepherds will guide the computer science and physics results, and that researchers should explain the math model output themselves.

Mathematicians on Kakeya and restriction, computer scientists who use the Exponential Time Hypothesis, and physicists in four-dimensional quantum field theory are the audience, along with readers tracking early access to an AI result. The report is part of All the drama around AI's takeover of mathematics. Math model output and a computer science Lean proof went on GitHub on Oct 26, 2026. The physics argument did not.

OpenAI three big science discoveries: the announcement

Local smoothing for the Euclidean wave equation is the math posting. OpenAI says an auxiliary partial differential equation establishes Sogge's local smoothing for waves in every spatial dimension at least two and yields the Kakeya maximal and dimension conjectures and Stein's spherical restriction estimates throughout the interior of the conjectured range. Model output is being released directly, as on October 6. A second description says the model produced a proof of the local smoothing conjecture in every dimension and therefore also the Kakeya conjecture and Stein's restriction conjecture in all dimensions.

The computer science paper is titled The Exponential Time Hypothesis is False. OpenAI says a deterministic algorithm exactly counts satisfying assignments of fixed-width CNF formulas in subexponential time and polynomial space, and that 3-SAT refutes that widely believed hypothesis. The physics result is not released in full. OpenAI calls it a mathematically rigorous existence proof of certain interacting quantum field theories with gauge fields and fermions in four dimensions, including Quantum Chromodynamics and a unified completion of the Standard Model.

What actually changed with OpenAI three big science discoveries

Mathematics follows the October 6 pattern of posting model output for researchers to explain. Computer science and physics use shepherds. UC San Diego's Russell Impagliazzo and Ramamohan Paturi, part of the group that proposed the Exponential Time Hypothesis in 1999, will check that result with MIT professor Ryan Williams. On Oct 26, 2026 they released the AI write-up and a Lean proof on GitHub. They asked OpenAI to make the files public immediately so others could inspect them with the group, not after a private first look.

OpenAI admits this approach could have its own issues. The day before, the AGMAI panel of math advisers said "we recommend that AI companies do not adopt" giving certain people the output first, citing unequal access, borrowed reputation, and biased selection. It wants scholarly repositories, Lean-formalized proofs, information on how problems were chosen, funding for people working on understanding the results, and other information. David Gross will lead the physics group: Michael Douglas, Arthur Jaffe, Igor Klebanov, and Nikita Nekrasov.

Who should care about OpenAI three big science discoveries

A few months ago, NYU professor Hong Wang received a Fields Medal from the International Mathematical Union for the Kakeya conjecture in three dimensions. OpenAI's manuscript claims Sogge's local smoothing in every spatial dimension at least two, the Kakeya maximal and dimension conjectures, and Stein's spherical restriction estimates inside the conjectured range. It also says the model resolves the local smoothing conjecture in every dimension and the Kakeya and Stein restriction conjectures in all dimensions. Mathematicians are comparing that broader claim with Wang's three-dimensional result.

Computer scientists who treat the Exponential Time Hypothesis as true are the audience for the 3-SAT result. OpenAI says it follows from a deterministic, subexponential, polynomial-space count of satisfying assignments of fixed-width CNF formulas. Impagliazzo, Paturi, and MIT's Ryan Williams are the examiners. Physicists on Quantum Chromodynamics and the Standard Model are the audience for the third claim, set in four dimensions with gauge fields and fermions, covering QCD and a unified completion of the Standard Model. That physics proof is unreleased.

How to try OpenAI three big science discoveries

On Oct 26, 2026 OpenAI put two packages on GitHub. Mathematics is model output titled Local smoothing for the Euclidean wave equation, released directly as on October 6. Computer science is the AI write-up and a Lean proof, public immediately at the request of an independent working group led by Impagliazzo, Paturi, and Williams. The claimed algorithm exactly counts satisfying assignments of fixed-width CNF formulas in subexponential time and polynomial space. OpenAI also says 3-SAT refutes the Exponential Time Hypothesis.

No full physics paper is posted. David Gross will shepherd the existence proof OpenAI described, on certain interacting quantum field theories with gauge fields and fermions in four dimensions, including QCD and a unified completion of the Standard Model, with Michael Douglas, Arthur Jaffe, Igor Klebanov, and Nikita Nekrasov. The computer science open letter asks what should govern early access, conflicts of interest, and unpublished material; how credit should cover a discovery plus checking, fixing, simplifying, explaining, and extending; and how to support students and researchers as methods change while preserving judgment to question results and develop new ideas.

What to watch after OpenAI three big science discoveries

Impagliazzo, Paturi, and Williams have not finished examining the claim that the Exponential Time Hypothesis is false. Gross, Douglas, Jaffe, Klebanov, and Nekrasov have not published the full four-dimensional proof on gauge fields, fermions, Quantum Chromodynamics, and a unified completion of the Standard Model. Other researchers still have to explain model output OpenAI posted directly, as on October 6, claiming local smoothing in every spatial dimension at least two and Kakeya and Stein results in all dimensions, past the three dimensions in Hong Wang's Fields Medal result.

The day before the Oct 26 posts, AGMAI said AI companies should not choose who sees output first. OpenAI says shepherds could have issues. Less than a month earlier it released hundreds of mathematics manuscripts. Robert Hart on Oct 6 reported another batch of mathematical breakthroughs. His Oct 9 headline, "Pure insanity," said mathematicians will need years to make sense of that drop. On Oct 15 he reported OpenAI is giving $5 million to help mathematicians make sense of AI-generated research.