OpenAI has released a large collection of mathematical research produced by an internal frontier AI model, including work on problems that have challenged professional mathematicians for years.
The company announced the release on October 6, saying the results cover a broad range of mathematics and are being published openly for scrutiny by researchers.
The public collection contains 722 manuscripts organized into 372 related result families, according to OpenAI's developer community. The work emerged from an evaluation involving roughly 4,000 mathematical problems.
The Verge reports that the collection spans hundreds of mathematical results generated by an unreleased frontier model. But publication does not mean every claim should immediately be treated as established mathematical fact. Researchers will need to examine the proofs and determine the importance and novelty of individual results.
One particularly significant feature is that OpenAI has provided formalizations of many proofs using Lean, a programming language that allows mathematical proofs to be checked by computer. The company says additional formalizations will be added over time.
OpenAI also revealed a striking measure of the computing involved. On average, each result required computing equivalent to roughly three hours of ChatGPT Pro thinking using the internal model.
The release follows increasingly dramatic claims about AI's ability to contribute to mathematics. In September, OpenAI said the same broader research effort had resolved more than 100 longstanding open problems, prompting the company to work with an independent group of prominent mathematicians on how such discoveries should be reviewed and communicated.
That advisory group includes mathematicians from institutions including Stanford, Harvard, Oxford, Cambridge, Berkeley and the Institute for Advanced Study. Its role is not simply to endorse OpenAI's work, but to independently advise on significance, research standards and responsible disclosure.
The larger question now goes beyond whether AI can answer difficult mathematics questions.
If independent mathematicians ultimately verify significant portions of this work, the new frontier may be whether AI can routinely discover mathematics that humans did not already know.
