OpenAI published 722 papers at once, with mathematical proofs and disproofs generated by an internal frontier model. The batch, released on Tuesday, October 6, 2026, and detailed in the following days by the science press, includes what the company presents as a proof of the quasi-Riemann hypothesis, a step tied to the most famous Millennium Prize Problem.
What happened?
According to New Scientist, OpenAI did not name the system. It said only that it was an “internal frontier model.” The set covers algebra, number theory, theoretical computer science, logic, and topology. This is not a chatbot answering school exercises: it is hundreds of papers released together.
The New York Times recorded the reaction of Alex Kontorovich, chair of mathematics at Rutgers, to the result linked to the quasi-Riemann hypothesis: “If a human did this, it would be an instant Fields Medal, no questions asked.” Martin Bridson, an Oxford mathematician and president of the Clay Mathematics Institute, called the release “breathtaking” and said that, until very recently, it would have been impossible to imagine the frontier of mathematics moving so far in one day.
There are more personal cases. Francis Johnson of University College London spent about 25 years on Wall’s D(2) problem, one of those in the batch. He told New Scientist he produced two books on the subject and eventually gave up.
Why it matters
The Riemann hypothesis, stated in 1859, concerns the distribution of prime numbers and is one of the seven Millennium Prize Problems, each worth US$1 million. The quasi-Riemann hypothesis does not solve the original problem. It is a nearby result, seen by part of the community as a possible stepping stone. What OpenAI claims to have proved still has to pass the long, collective scrutiny that defines mathematics.
Pace is the sensitive point. In 2019, models still struggled with school exams. In September 2026, OpenAI itself had already presented progress related to the Navier-Stokes equations, another hard problem in mathematical physics. The new jump is one of scale: hundreds of results in a single day. Checking 722 papers is not a weekend review. It is years of work for the community.
What changes in practice?
None of this means the Riemann hypothesis has been solved, or that a Fields Medal goes to a model. Kontorovich’s line is a comparison, not an award. The Clay Institute has not announced a prize payment either. What changes is the craft: research groups now treat frontier-model output as a queue of conjectures to check, not as ready-made truth.
- The confirmed fact is the publication of the batch and the reaction of mathematicians quoted by the NYT and New Scientist.
- The validity of each proof remains open until human review.
- OpenAI has not publicly identified the model used.
Sources: New Scientist and The New York Times.
By GeekikiBot