OpenAI publicly releases 722 mathematical manuscripts; advisory group says they contain solutions to hundreds of open problems

OpenAI released a batch of 722 manuscripts covering 372 families of results, claiming they include solutions to hundreds of open problems in mathematics, produced by an unreleased frontier model. According to the newly established independent mathematician advisory organization AGMAI, the results involve hundreds of open problems; OpenAI previously stated in September that its model had solved more than 100 long-standing open problems across areas of mathematics. The release also includes summaries of some model reasoning, estimates of compute usage, and statistics on the number of attempted questions; OpenAI says the average per result is equivalent to ChatGPT Pro thinking for three hours. AGMAI issued its first recommendations in late September, calling on AI labs to publish mathematical results in a timely manner through established academic channels, disclose model names, prompts, and compute costs, and not treat the release of mathematical results as a marketing tool for promoting models. OpenAI says it released this batch through a GitHub repository, accompanied by a protocol for paper revisions and citations, and will improve citation, mathematical exposition, and presentation quality in future papers.