OpenAI has released 722 research manuscripts organized into 372 result families, covering areas such as number theory, algebraic geometry, and theoretical computer science. These findings were generated by an internal, unreleased AI model that was tested against approximately 4,000 problems.
As of early October 2026, 42% of the results, totaling about 300 items, include Lean formalizations. Lean is a proof assistant software that allows for machine-based verification of logical steps. Three manuscripts were retracted shortly after publication due to identified errors.
The project was guided by an advisory group from the Institute for Advanced Study, incorporating feedback from hundreds of mathematicians. While the release touches on Millennium Prize Problems, including the Riemann hypothesis, OpenAI has not claimed to have solved any of them.
In September 2026, OpenAI reported progress on the Navier-Stokes equations using approximately 10,000 AI agents over an 88-hour period. Critics have raised concerns regarding the verifiability of the new results, as the underlying model and specific prompts remain undisclosed.
