OpenAI Releases Solutions to Hundreds of Mathematical Problems via AI Model

OpenAI has made public a substantial collection of mathematical findings generated by an unreleased frontier model, with a repository of 722 manuscripts covering 372 groups of related papers now available for the research community to examine. The collection represents resolutions to what's described as "hundreds" of previously unsolved mathematical questions, according to AGMAI, an independent panel of respected mathematicians assembled to guide responsible disclosure of such work.

What was released#

The publication includes more than 100 long-standing problems resolved across most mathematical domains, along with supporting materials that give insight into how the model arrived at its answers. The researchers provided summaries of the model's reasoning process, calculations of computational resources required—the company says an average solution consumed roughly three hours of ChatGPT Pro thinking time—and metrics showing how many problems were attempted overall.

The governance framework#

AGMAI has been instrumental in shaping this announcement. The advisory group issued recommendations in late September calling for AI laboratories to share mathematical discoveries through traditional academic publishing channels whenever feasible, while being transparent about critical details: the specific model name, the prompts used, and the computational cost. Critically, AGMAI urged companies to stop treating mathematical breakthroughs as marketing opportunities to boost their brand, arguing that this approach causes genuine damage to the mathematics community's research culture.

How OpenAI approached the release#

The company published its findings via GitHub and established protocols for updating papers and managing citations. OpenAI has indicated it will continue exploring community-managed publishing options that align with AGMAI's recommendations, and committed to improving the quality of future releases by refining how papers cite prior work, explain mathematical concepts, and present findings for broader comprehension.

Ongoing tensions#

This release is the latest contribution to what has become a rapidly expanding collection of AI-derived mathematical results from OpenAI and competitors like Anthropic. The acceleration of AI breakthroughs in mathematics this year—particularly how these discoveries are announced—has triggered significant friction within the academic community. Mathematicians are grappling with questions around research integrity, whether AI companies properly acknowledge the human mathematicians whose prior work trained these systems, and whether such systems' results should count as novel contributions to the field.