OpenAI posts 722 math papers from an unreleased model, including a claimed quasi-Riemann hypothesis proof
One entry claims every Dirichlet L-function, zeta included, has no zeros past 7/8, and the model behind it has no name.
On October 6, 2026, OpenAI posted a large batch of new mathematics produced by a model it hasn't released. The results sit in a public GitHub repository: 722 manuscripts, grouped into 372 families of related results, with an announcement post on OpenAI's site.
OpenAI calls the source of all this "an internal frontier model" and doesn't give it a name. It says it's working to release that model responsibly, and gives no date.
- Who
- OpenAI
- What
- 722 manuscripts covering 372 result families
- Problems posed
- approximately 4,000
- Compute per result
- three hours of ChatGPT Pro thinking, on average
- Checked by computer
- Lean formalizations for some, not all
- Model
- unreleased
What's in the repository
Each family groups a main result with the papers around it. The repository README says the model was posed approximately 4,000 problems, and the manuscript map lists what came out.
The entry that drew the loudest response is number 003, on the quasi-Riemann hypothesis. The map says the model's paper proves every Dirichlet L-function has no zeros to the right of the line where the real part is 7/8, with this conclusion:
In particular, the Riemann zeta function is zero-free in this half-plane, proving the quasi-Riemann hypothesis.
There's a second, weaker proof of the same family with a cutoff of 11/12, and the README says that writeup "was human edited for readability." Entry 074 covers Kakeya problems, the maximal conjecture in three dimensions and the dimension conjecture in four.
Some of the proofs come with Lean formalizations, and OpenAI says it'll add more as it gets them. Ten results come with abridged summaries of the model's reasoning.
How OpenAI released it
OpenAI says it consulted the Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study. That group published recommendations on September 29, 2026, a week before the drop.
make public the name of the model, the prompts used, a (summarized) chain of thought, the time taken, and the estimated cost of computation
OpenAI covered some of that list: compute estimates, reasoning summaries and attempt counts. It didn't name the model. The group also wants results in scholarly repositories that AI labs don't control, and OpenAI says it's still looking at "community-hosted alternatives" for this release. For now, it's GitHub.
It also says it'll fund workshops, conferences and special programs on understanding results produced by AI, with details to come.
What mathematicians said
Mathematicians on X answered the same evening. Daniel Litt spotted a link to his own work, and to a student's.
Fun! Looks like mathematicians have a lot of exciting work to do. If I understand correctly one of the results is a (very special) case of a conjecture of mine, which is also a consequence of stronger work in progress by a student of mine.
Alex Kontorovich went straight to entry 003.
Quasi-RH?!?!???! Are you kidding me? If a human did this, it would be an instant Fields Medal, no questions asked. RH says zeta has no zeros in Re(s)>1/2. The best we had until a second ago was a region that got thinner and thinner the higher up the imaginary axis you go. I thought maybe they’d fatten that up a bit, that’d be a massive breakthrough. But no. They got a zero free strip!!!! Insane
Not everyone was celebrating. Alvaro Lozano-Robledo asked what releasing hundreds of results at once is meant to achieve.
The scale of this is staggering. But... Why? Why is OpenAI trying to solve hundreds of math problems to be released all at once? I think we got the point when they found a counterexample to Navier-Stokes -- internal model = great. I am not sure at this point what this (hostile?) takeover of the mathematical landscape is trying to achieve other than, again, a massive PR stunt.
The README says OpenAI will preserve the collection's public release history, with corrections and revisions logged in the repository. The two quasi-Riemann papers carry the dates September 30 and October 5, 2026 in their folder names.
More on OpenAI
- Common Sense Media rates ChatGPT for Teens an Unacceptable Risk and wants it 18-plus until fixedOctober 7, 2026
- Sam Altman says the world should accept some bad things from AI, and senators push backOctober 6, 2026
- OpenAI will watermark ChatGPT and Codex text in the EU, and says rewriting can remove itOctober 5, 2026
- Wikimedia says OpenAI's agents edited its wikis and may have helped cause a May outageOctober 5, 2026