OpenAI announced that an unpublished frontier model has solved numerous enduring mathematical problems, presented in 722 manuscripts that encompass 372 families of related results. This continues a series of discoveries that have both astonished and concerned sections of the math community, prompting ethical and conduct debates.
As noted by AGMAI, the independent panel of top mathematicians assembled to ensure responsible communication, the batch contains solutions to “hundreds” of unresolved questions. It is claimed that the batch contains solutions to “hundreds” of open mathematical questions.
The findings were anticipated for weeks, but OpenAI had not yet disclosed which problems were tackled or the exact publication timeline.
In September, the firm reported that its model had “solved more than 100 long‑standing open problems spanning most mathematical domains.” The accompanying documents also feature brief summaries of the model’s reasoning, compute‑usage estimates, and statistics on the total problems tried. OpenAI states that the typical outcome required roughly three hours of ChatGPT Pro‑level computation.
AGMAI — the Advisory Group on Mathematics and Artificial Intelligence — issued its initial guidance in late September, urging AI laboratories to publish mathematical results promptly and via conventional academic channels when feasible, while also revealing information such as the model name, prompts, and compute expenses.
The group further urged AI companies to avoid using the release of mathematical results as promotional tools for their models, noting that this harms the mathematical community. OpenAI outlined its approach for this release: the results are being posted in a GitHub repository, which includes procedures for revising papers and citing them.
For this release, we’re publishing the results in a GitHub repository, with protocols for paper revisions and citations.