Editorial illustration for OpenAI releases 372 AI-generated math proofs, funds academic workshops

OpenAI releases 372 AI-generated math proofs, funds...

By AI Daily Post Edited by Brian Petersen, Editor-in-Chief

October 7, 2026 • 3 min read

OpenAI put 372 mathematical results on GitHub this week, not in a peer-reviewed journal, and billed them as solutions or serious progress on open problems in the field. Some touch improvements to computer algorithms that run far beyond academia. Others chip away at the Riemann hypothesis, a problem mathematicians have chased since 1859. The company is hosting the whole batch in a public repository, with revision logs and citations attached, as if inviting outsiders to check the work themselves rather than wait for a journal to referee it.

The timing matters. OpenAI says the same internal model behind this release already produced a Navier-Stokes solution that has sat under formal review for weeks, a problem serious enough to carry a Clay Millennium Prize. That earlier result took a coordinated swarm of 10,000 agents and millions of dollars in compute to pull off. This new batch of 372 results didn't need anywhere close to that firepower, which raises its own questions about what changed between the two efforts and what it means for how fast these proofs can now be generated.

OpenAI has published 372 new mathematical results generated by an internal frontier model. Each result is supposed to solve an open problem or make substantial progress toward one. The collection includes improvements to major computer algorithms and advances related to the Riemann hypothesis.

OpenAI dumps 372 AI-generated math proofs on GitHub, telling the academic world to keep up - THE DECODER

Why this matters

The 372 results sound impressive until you notice the fine print: most came from a single prompt run through a single agent, chewing through roughly three hours of ChatGPT Pro compute each. That's not a research program, it's a very expensive brute-force search, and OpenAI is asking mathematicians to validate the output after the fact. Funding workshops to help academics make sense of AI-generated proofs is a tell. If the work were clean, it wouldn't need an interpretive layer.

For researchers, the practical question is verification cost. A proof that takes three GPU-hours to generate but weeks of human expert time to check isn't obviously a time-saver, especially when OpenAI itself admits the citations and presentation need work. For founders building on top of these models, the lesson is similar: generating plausible-looking output at scale is the easy part.

Getting domain experts to trust it is the actual bottleneck, and no amount of GitHub volume substitutes for that. Watch whether any of these 372 results survive peer scrutiny intact.