OpenAI shares new math and physics results with academic researchers
OpenAI said on 2026-10-26 that it is sharing model results on several research problems in mathematics and physics and outlining how outside researchers can verify and build on the work. The announcement adds to a run of recent research disclosures from the company, but the source material makes clear that the reported results should be treated as company-reported findings unless and until independent experts confirm them.
Evidence and context
In the 2026-10-26 post, OpenAI said it is publishing results related to the Kakeya conjecture, the Exponential Time Hypothesis, and quantum chromodynamics. The company also described how researchers can verify and extend the outputs. According to the enrichment data, OpenAI says a claimed local-smoothing proof was generated on 2026-10-18 and formalized on 2026-10-19, while a claimed disproof of the Exponential Time Hypothesis was generated on 2026-10-16 and formalized on 2026-10-17. Those dates are company-reported and refer to model output and formalization, not independent acceptance.
The source material also says the physics collaboration concerns existence results, including pure Yang–Mills existence. It should not be read as an announcement that the full Yang–Mills existence-and-mass-gap Millennium Prize problem has been solved. The enrichment notes that OpenAI initially withheld the AI-produced artifacts for the physics work, underscoring that the release status differs across the research areas discussed in the post.
The announcement follows earlier OpenAI research disclosures cited in the enrichment data: on 2026-09-21, the company said it had engaged an independent mathematics advisory group to improve review, dissemination and academic standards; on 2026-10-06, it published mathematical outputs and computer-checkable Lean formalizations alongside details of its research process; and on 2026-10-21, it posted about using thousands of AI agents to uncover new leads in biological data. Together, those posts show an active effort to frame model outputs in ways that academic communities can inspect, formalize and test.
The available evidence remains limited to OpenAI's own announcements and the supplied context. The material does not establish that the mathematical claims have been independently verified, nor does it show experimental validation of the physics results beyond the company’s description of the work. No next verified catalyst date was supplied.