OpenAI researchers tested an unreleased AI model by assigning it 4,000 previously unsolved mathematical problems. The experiment resulted in the creation of 722 manuscripts, which the company organized into 372 families of related findings.
According to OpenAI, each generated result required an average of three hours of reasoning compute. The company noted that while many of these papers include computer-checkable Lean proofs, others remain unverified and could contain errors.
To manage the research process, OpenAI consulted with the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study. The company officially announced the release of these findings on October 6, 2026.
The experiment highlights a shift in AI capabilities from answering known questions to generating potential solutions for unknown ones. This development raises concerns regarding the human capacity to verify the volume of research produced by AI systems.
