OpenAI Solves 722 Math Problems and Proves Quasi-Riemann Hypothesis in Breakthrough

A sudden mathematics release with an unusually large scope

OpenAI’s latest mathematics announcement is not a single paper but a large public release of AI-produced research material. On October 6, 2026, the company said it was releasing “a broad range of new mathematical results” generated by an internal frontier model, and placed the materials in a public GitHub repository with citation and revision protocols [1]. The repository describes the package as mathematical manuscripts and supporting proof artifacts produced by an internal OpenAI model [2].

The headline number is 722 manuscripts, organized into 372 “families” of related results [2]. That distinction matters. OpenAI is not saying there are 722 unrelated theorems; a family can include a main paper, companion arguments, consequences, or alternative proofs [2]. Still, the scale is extraordinary: even if only a fraction of the catalogue survives review, the release could become one of the most consequential demonstrations yet of AI systems doing frontier mathematical work.

The most attention-grabbing entry is Family 003, titled “The quasi-Riemann hypothesis.” OpenAI’s overview states that the result proves every Dirichlet L-function, including the Riemann zeta function, is zero-free in the half-plane Re s > 7/8, and says this resolves the quasi-Riemann hypothesis [3]. The same catalogue entry also says the same half-plane is zero-free for finite-order Hecke L-functions over Q(√−3), and lists a companion proof for the weaker Re s > 11/12 region [3].

What OpenAI says the model did

According to OpenAI, the results came from an internal frontier model rather than a currently released public product [1]. The company says it expanded its evaluation program after existing mathematical benchmarks became saturated, then posed roughly 4,000 open research problems to the model [2]. After aggregation into significant result families and manuscripts, the public catalogue contained 722 manuscripts across 372 families [2].

OpenAI also gave a rare estimate of compute. The repository says the vast majority of results were obtained with the same procedure and used, on average, the equivalent of three hours of ChatGPT Pro thinking compute per result [2]. OpenAI’s announcement likewise says the release includes compute estimates, statistics about attempted problems, and ten abridged summaries of the model’s reasoning [1].

This is part of why the release is being read as a claim about both mathematics and scientific process. If a model can repeatedly produce serious manuscripts from open research prompts, the bottleneck shifts from generation to verification. That is precisely where OpenAI’s own caveats begin.

The quasi-Riemann claim, explained without overstating it

The full Riemann Hypothesis predicts that the nontrivial zeros of the zeta function lie on the critical line Re s = 1/2. OpenAI’s claim is not that full theorem. The quasi-Riemann claim described in the catalogue is a zero-free half-plane: no zeros for Dirichlet L-functions in Re s > 7/8 [3]. In plain language, the proposed result would push a hard boundary away from 1 and toward the critical line, a major strengthening in analytic number theory if confirmed.

The catalogue’s wording is strong: it says the entry “proves” the zero-free region and “resolves” the quasi-Riemann hypothesis [3]. However, the current state is still a release of manuscripts and proof artifacts, not a completed external refereeing process. OpenAI’s repository says the collection includes results at different stages of verification, not all of them have Lean formalizations, and some unformalized results could have issues [2]. Independent analysis published after the release has therefore described the collection more cautiously as a set of major claims with uneven evidence, rather than a set of universally accepted theorems [5].

That distinction is not nitpicking. In mathematics, a proof becomes part of the accepted record through expert scrutiny, correction, formal checking where possible, and ultimately community confidence. The OpenAI release accelerates the first step by making papers and artifacts public, but it also compresses years of potential reading into a single drop.

Lean formalization and the verification question

A key feature of the announcement is the use of Lean, a proof assistant that allows mathematical statements and proofs to be checked by machine. OpenAI said it is sharing formalizations of many proofs in Lean and will update the repository as more become available [1]. The README says many, but not all, manuscripts have been formalized, and points readers to a Lean library and formalization catalogue for checking configurations [2].

Outside summaries of the release have emphasized the same caveat: some results have formal proof artifacts, while others still rely on manuscripts alone [4]. AICoder’s technical summary reported that many families link to Lean materials and that the formalization catalogue lists a subset of papers with formalized main results, while also noting that nothing has yet gone through peer review [4]. Kingy AI similarly stressed that proof artifacts make the release more inspectable, but do not automatically convert every manuscript into an independently accepted result [5].

This makes the quasi-Riemann entry especially important. The mathematics community will want to know not only whether the argument is correct, but also what parts, if any, are formalized; whether the formal statements match the advertised theorem; and whether all dependencies are acceptable. A machine-checked proof can be powerful evidence, but only if the formal statement captures the intended mathematical claim.

Why the release is already causing debate

The release immediately generated intense discussion because it combines three volatile ingredients: famous open problems, an unreleased model, and an enormous verification burden. HuggingNews summarized the quasi-Riemann result as a claimed proof with a 7/8 bound and noted that the broader release includes claimed progress on several major mathematical fronts [6]. The same report also highlighted that the model remains unreleased, meaning outside mathematicians can inspect the outputs but cannot reproduce the generation process itself [6].

OpenAI says it consulted the Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study and drew on that group’s public recommendations in deciding how to release the results [1]. The company also says it will fund workshops, conferences, and special programs focused on understanding major AI-produced results [1]. That is a recognition that publication alone is not enough. Hundreds of advanced manuscripts require a social and institutional response: experts need time, incentives, and shared standards for triage.

The catalogue’s size creates a new editorial problem for mathematics. A traditional breakthrough might be a single preprint that a small group of specialists can study in depth. Here, the community faces hundreds of documents spread across number theory, algebraic geometry, theoretical computer science, combinatorics, mathematical physics, operator algebras, and more [5]. Verification becomes an allocation problem: which claims should be checked first, who is qualified to check them, and how corrections should be tracked.

The strongest reading and the cautious reading

The strongest reading is that OpenAI has demonstrated a major leap in AI-assisted mathematical discovery. Under that interpretation, the quasi-Riemann proof is the flagship result: a model-generated advance in analytic number theory that would have been headline news even as a standalone human-authored paper. The wider 722-manuscript catalogue then becomes evidence that frontier models are no longer merely solving contest-style problems but producing research programs.

The cautious reading is that OpenAI has released an enormous collection of proposed results, some supported by formal artifacts and some not, and that the true breakthrough will only be known after expert review. OpenAI’s own documentation supports this caution by saying results are at different stages of verification and that unformalized results could contain issues [2]. Independent coverage has also emphasized that the manuscripts are not peer reviewed and that the release should be treated as a catalogue of claims until the mathematics is checked [4][5].

Both readings can be true at once. The act of producing 722 serious-looking manuscripts is already a meaningful event in AI for science. The act of proving the quasi-Riemann hypothesis, however, is a mathematical claim that must meet the field’s highest standards.

What happens next

The next phase will be less spectacular than the announcement but more important. Mathematicians will examine the quasi-Riemann manuscripts, compare the 7/8 proof with the 11/12 companion proof, test the Lean artifacts where available, and look for hidden assumptions, gaps, or mismatches between formal and informal statements. OpenAI says it will preserve release history and record corrections and revisions as new versions, leaving earlier versions accessible [2].

If the quasi-Riemann proof survives that process, the October 6 release will be remembered as a landmark in both number theory and AI research. If it fails, the release will still matter as a case study in how powerful models can generate plausible, ambitious mathematics faster than humans can certify it. Either way, OpenAI has forced a practical question onto the field: how should mathematics handle discovery when the production of conjectures, proofs, and manuscripts becomes faster than the traditional machinery of review?

For now, the accurate description is this: OpenAI has announced a massive AI-generated mathematics release, including 722 manuscripts and a claimed proof of the quasi-Riemann hypothesis. The breakthrough may be real, but its final status will be decided not by the announcement, and not by the headline, but by verification.

Developments

1. OpenAI Solves 722 Math Problems, Proves Quasi-Riemann Hypothesis36Kr · Oct 7, 2026, 4:24 AM · 9/10
2. OpenAI releases 722 AI-generated math manuscripts claiming breakthroughsBigGo Finance · Oct 7, 2026, 3:55 AM · 8/10
3. OpenAI Releases 722 AI-Generated Math Manuscripts Claiming BreakthroughsBigGo Finance · Oct 7, 2026, 3:55 AM · 8/10
4. OpenAI Releases 722 Manuscripts with Mathematical Results and CostsThe New York Times · Oct 7, 2026, 1:46 AM · 8/10
5. OpenAI releases 722 manuscripts with mathematical results — The Verge - UA.NEWSUA.NEWS · Oct 7, 2026, 1:40 AM · 8/10
6. OpenAI’s 722 Math Manuscripts: The Results, Proofs, Compute and Costs - Kingy AIKingy AI · Oct 7, 2026, 1:28 AM · 8/10
7. OpenAI Releases 722 Math Manuscripts from Unreleased AI ModelUnite.AI · Oct 7, 2026, 12:39 AM · 9/10
8. OpenAI releases 722 math manuscripts from unreleased AI modelUnite.AI · Oct 7, 2026, 12:39 AM · 7/10

Sources from the last 72 hours

1. [1]Sharing AI progress in mathematicsOct 6, 2026, 2:00 AM
2. [2]math/README.md at main · openai/math · GitHubOct 6, 2026, 2:00 AM
3. [3]math/overview.tex at main · openai/math · GitHubOct 6, 2026, 2:00 AM
4. [4]OpenAI publishes 722 math manuscripts from an unreleased internal frontier model on GitHub, claiming results on the quasi-Riemann hypothesis, the Hodge conjecture for CM abelian varieties and moreOct 7, 2026, 2:00 AM
5. [5]OpenAI’s 722 Math Manuscripts: The Results, Proofs, Compute and CostsOct 6, 2026, 2:00 AM
6. [6]OpenAI Math Release Includes Claimed Progress on Three Millennium Prize ProblemsOct 7, 2026, 12:32 AM

AI-generated article based on recent web research, then preserved as a dated editorial snapshot.