OpenAI Releases 722 Papers Tied To Unsolved Math Problems
OpenAI is routing AI-generated mathematics through a review and citation process built with an outside advisory group before publication, which is the part of this release that sets a precedent for how the results get absorbed by mathematicians.
Reporting from 1 source: GIGAZINE.
OpenAI published a GitHub repository on October 6, 2026 holding 722 papers across 372 result families, described as new mathematical results from its in-house models. Many proofs carry Lean formalizations, a language used to make them machine-checkable. OpenAI said the release follows discussions with the advisory group AGMAI and its recommendation on responsible publication of AI-generated mathematical information. The repository will be updated as new formalizations arrive.
The repository sorts the manuscripts by mathematical field and includes proof materials. OpenAI also published procedures for revision and citation alongside the results and computational procedures. For many papers, Lean formalizations accompany the proofs so a computer can verify them, and OpenAI said it plans to keep updating the repository as new formalizations become available.
Outside readers marked specific items. Researcher Isaac Kim, whose work covers quantum information and many-body physics, described the list as containing shocking problems in his areas and said things are changing rapidly. OpenAI said the release follows discussions with AGMAI and its recommendation on responsible publication of AI-generated mathematical information, and that it is working toward releasing the models responsible in a responsible manner.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.