🔍 Read the full analysis: OpenAI’s AI Mathematics: Where Might 722 Proofs Take Us? on ThorstenMeyerAI.com

Prime Big Deal Days · Oct 6–7Offer from Amazon

Get tech for your team delivered free — and shop member deals

• Fast, free delivery on millions of items
• Access to Prime Big Deal Days deals on October 6–7
• Prime Video, Amazon Music and more included

Start your free Prime trial →Free trial for eligible customers · Cancel anytime

As an affiliate, we earn on qualifying purchases.Hide

OpenAI published 722 mathematical manuscripts on Monday, presenting work by a model the company has not released or named. The papers, arranged into 372 families of related results, include claims about several major open problems, but OpenAI chief executive Sam Altman said the claims have not been confirmed by outside mathematicians.

OpenAI’s post and repository describe results across number theory, geometry, topology, operator algebras, theoretical computer science and mathematical physics. The manuscripts came from roughly 4,000 problems posed to the model. OpenAI says it filtered that pool for problems it considered significant, meaning the company—not independent reviewers—made the selection. The average result used about three hours of ChatGPT Pro thinking compute, according to the source material.

The catalogue includes asserted proofs or resolutions concerning the Unique Games Conjecture, Hilbert’s tenth problem over the rationals, the Hodge conjecture for CM abelian varieties, and a longstanding question about nonabelian free group factors. Another manuscript claims a zero-free region for the Riemann zeta function to the right of Re(s) = 11/12. These are claims in the published manuscripts, not established mathematical breakthroughs.

OpenAI released Lean formalizations for many, but not all, results. Its repository warns that some unformalized results could have issues. The release also provides only ten abridged reasoning summaries for the 372 families. The source material says the Riemann manuscript was edited by humans for readability; it does not establish that the claimed results have passed independent review.

Verification Will Shape the Payoff

The release’s importance will depend on more than whether a theorem’s statement is true. Mathematicians often value a proof for the methods and ideas it makes available, not just the problem it settles. If researchers can verify the arguments, extract techniques and apply them elsewhere, the work could support further discoveries. If the proof is correct but difficult to interpret or reuse, its effect may be narrower.

The Unique Games Conjecture illustrates the potential stakes: many results in theoretical computer science rely on assumptions connected to it, including claims about the limits of approximation algorithms. A valid proof could prompt researchers to revisit those arguments. But until specialists verify the manuscript and clarify precisely what it establishes, the practical consequences remain conditional, not confirmed.

The release also tests how AI-generated mathematics should be evaluated. Formal verification can check that a Lean proof follows from its formal definitions and assumptions, but formalization is not available for every manuscript, and it does not by itself show that a result is useful or that the formal statement matches the intended mathematical claim. Human review remains necessary to establish both correctness and significance.

OpenAI’s Recent Math Releases

This is OpenAI’s fourth major mathematics release this year, according to the source material. In May, its model produced a counterexample to the Erdős unit-distance conjecture. Five mathematicians later published what they called a digested, human-verified version of the work. That process—turning machine output into an argument people can evaluate—offers one possible route from a generated result to accepted mathematics.

An August release, called “Ten Advances,” had a more contested reception. A claimed counterexample to Connes’s rigidity conjecture was challenged within a day; a critique argued that the groups constructed did not meet a condition required by the conjecture. In September, OpenAI announced a Lean-formalized Navier–Stokes result produced by about 10,000 concurrent agents over 88 hours. That announcement prompted a dispute over priority and concern among mathematicians about using famous problems as benchmarks without sufficient human understanding.

The debate is not simply whether AI can produce a correct proof. A computer-assisted proof such as the Four Colour Theorem settled a question, but its case-checking approach offered less reusable theory than proofs whose methods opened new avenues of research. The current collection will need to be judged result by result: some claims may fail, some may be correct but hard to build on, and others may yield ideas researchers can use.

“Digested, human-verified.”

— The five mathematicians who reviewed the Erdős unit-distance result

Independent Checks Still Needed

No outside confirmation is reported for the 722 manuscripts as a collection. The source material does not identify independent reviewers who have verified the headline claims, nor does it say how many manuscripts have been checked in detail. It also does not provide enough information to assess the model’s identity, training, or full reasoning process.

Formal proof files cover many results, but not all; ten abridged summaries represent only a small portion of the 372 families. The details of OpenAI’s selection process and the criteria used to judge significance are also limited. It remains unclear which claims will withstand scrutiny, whether any proof depends on assumptions or definitions that differ from the original problem, and which results will produce reusable mathematical ideas.

Mathematicians Must Test the Claims

The next step is independent examination of the manuscripts and available formalizations. Researchers will need to check whether each argument is valid, whether it proves the stated problem, and whether its methods can be explained and reused. The Erdős example suggests that a readable, human-verified account may help the field assess a result; the earlier dispute over Connes shows why claims should not be treated as settled before that work is done.

OpenAI has not, in the source material, set out a timetable for external review or announced a complete set of expanded reasoning notes. For now, the 722 papers are a large collection of AI-attributed mathematical claims, not 722 accepted discoveries. Their significance will become clearer as mathematicians publish verification, corrections or objections—and show whether any proof leads to further work.

Key Questions

What did OpenAI release?

OpenAI published 722 mathematical manuscripts, grouped into 372 families, based on work by an unnamed, unreleased model. The company says the work followed roughly 4,000 posed problems.

Have the claimed proofs been verified?

Not as a collection. The source material says the claims have not been confirmed by outside mathematicians, and OpenAI’s repository cautions that some unformalized results could have issues.

What major problems do the manuscripts address?

Among the claims are results concerning the Unique Games Conjecture, Hilbert’s tenth problem over the rationals, the Hodge conjecture for CM abelian varieties and the Riemann zeta function. These remain claims awaiting independent scrutiny.

Why does it matter whether the proofs are understandable?

A proof can settle a question yet contribute little reusable knowledge. Researchers will assess not only whether the arguments work, but whether their methods can be understood and applied to other problems.

What happens next?

Mathematicians will need to examine the papers and formalizations, identify errors or confirm results, and determine whether the arguments offer useful techniques. The source material gives no timetable for completing that review.

Source: ThorstenMeyerAI.com