Latest

The mathematics frontier · Vol. 01 / OpenAI

New proofs.
Bigger questions.

AI is reaching into mathematics that has challenged researchers for years. Explore the claims, understand the ideas, and follow the evidence behind each result.

From an idea to shared knowledge

  1. 01

    Generate

    A claimed result

  2. 02

    Check

    Proof and scope

  3. 03

    Understand

    Human insight

Conceptual illustration, not a measured progress curve.

manuscripts indexed
719
result families
372
mathematical fields
17
plain-language guides
20

Start with an idea you can picture

Three doors into the collection

Browse the research

The proof directory

How to read the evidence

Browse by result family, then open its individual manuscripts. Plain-language guides are marked; other entries show OpenAI's source summary. All results are attributed claims.

372 result families · 719 manuscripts · Page 1 of 21

Number theory#001

Milne’s rationality conjecture and algebraic specialization

OpenAI source summary

Proves Milne's rationality conjecture for abelian varieties over \overline{\mathbb Q} with good reduction: specialized Hodge classes pair rationally with complementary divisor products, independently of cohomology theory. Together with result 032, every specialized Hodge class is represented by a single rational algebraic cycle simultaneously in all prime-to-p and crystalline realizations, for every residue characteristic p.

1 manuscriptNo Lean link listed
Number theory#002

The full BSD formula from low Selmer corank

OpenAI source summary

Proves the full Birch–Swinnerton-Dyer leading-term formula for every elliptic curve over ℚ whose full q-power Selmer group has corank zero or one for some prime q, including finiteness of the Tate–Shafarevich group. With result 006, this gives full BSD for a density-one set of quadratic twists of every elliptic curve over ℚ.

3 manuscriptsNo Lean link listed
Number theory#003

The quasi-Riemann hypothesis

OpenAI source summary

Proves that every Dirichlet L-function, including \zeta(s), is zero-free in \Re s> 7/8, resolving the quasi-Riemann hypothesis. The same half-plane is zero-free for every finite-order Hecke L-function over \mathbb Q(\sqrt{-3}). A companion gives a different proof of the zero-free half-plane \Re s> 11/12.

3 manuscriptsLean scope linked
Number theory#007

A hidden randomness in prime factors

Plain-language guide

Count whether each integer has an even or odd number of prime factors, including repeats. This paper claims a precise cancellation result for paired patterns in that sequence, together with a wider result for related arithmetic functions.

1 manuscriptLean scope linked
Number theory#008

The Deligne–Drinfeld conjecture

OpenAI source summary

Proves that the rational Grothendieck–Teichmüller Lie algebra, with the Ihara bracket, is freely generated by one element in each odd weight 3, 5, 7, …, resolving the Deligne–Drinfeld conjecture.

1 manuscriptLean scope linked
Number theory#009

Function-field reconstruction from Milnor K-theory and Galois data

OpenAI source summary

Reconstructs function fields of transcendence degree at least two over algebraically closed constants from K^{\mathrm M}_1/\ell, K^{\mathrm M}_2/\ell, and their product. These data recover the perfect closure and constants when ℓ differs from the characteristic, and the original field and its named base in equal characteristic. Also proves Bogomolov–Pop reconstruction from abelian-by-central pro-ℓ Galois data away from the characteristic.

3 manuscriptsLean scope linked
Number theory#010

Unrestricted pro-modularity at the prime two

OpenAI source summary

Every continuous odd absolutely irreducible two-dimensional 2-adic representation of G_{\mathbb Q} unramified outside finitely many primes occurs in a completed Hecke algebra at some odd tame level. Also proves classical modularity up to Tate twist for irreducible odd representations with these finiteness conditions that are de Rham at 2 with distinct Hodge–Tate weights, resolving the dyadic Fontaine–Mazur case without residual restrictions.

3 manuscriptsNo Lean link listed
Number theory#011

Prime-factor statistics of p-1

OpenAI source summary

Proves that the normalized ordered logarithms of the prime factors of p-1, counted with multiplicity, converge jointly to the Poisson–Dirichlet law PD(1) as p ranges uniformly over primes up to x and x→∞. This resolves the Ford–Konyagin–Luca conjecture. It also proves that infinitely many integers n have more than n^{1-ε} totient preimages, for every ε > 0.

3 manuscriptsNo Lean link listed
Number theory#014

Restricted geometric Langlands, global Arthur enhancements, and generic Ramanujan

OpenAI source summary

Proves the restricted geometric Langlands equivalence for connected reductive groups on smooth projective connected curves over \overline{\mathbb F}_q under the four stated Lie-theoretic characteristic hypotheses. Over arbitrary algebraically closed fields of characteristic p > 0, the same conclusion holds assuming additionally that p is very good for the group and p\nmid |W_G|. Over global function fields, proves Ramanujan at every place for globally generic cuspidal representations of split adjoint absolutely simple exceptional groups, without characteristic or ramification-depth restrictions, and at every unramified place for cuspidal representations of split adjoint absolutely simple groups with a generic unramified component. Assuming the finite-level Ramanujan–Arthur decomposition, constructs global Arthur enhancements of occurring cuspidal excursion parameters for split connected semisimple groups at full finite level, recovering the given parameters by diagonal specialization on the entire Weil group, including inertia.

8 manuscriptsNo Lean link listed
Number theory#015

Torus-packet equidistribution in prime, quartic, and sextic degrees

OpenAI source summary

Proves Haar equidistribution without escape of mass for complete volume-weighted torus packets from totally real fields: arbitrary lattices and prescribed local types in fixed prime degree at least five, arbitrary-order Picard packets in primitive quartic fields, and maximal-order ideal-class packets in primitive sextic fields. Here primitive means having no proper intermediate field; the relevant order or field discriminant tends to infinity.

3 manuscriptsLean scope linked
Number theory#016

Zilber–Pink in abelian varieties and the Siegel threefold

OpenAI source summary

Proves the abelian Zilber–Pink conjecture over \overline{\mathbb Q}: every irreducible subvariety has finitely many maximal atypical subvarieties relative to its smallest containing torsion coset. It also proves the full curve case in the Siegel threefold \mathcal A_2 for Hodge-generic curves defined over \overline{\mathbb Q}, without boundary or reduction assumptions.

4 manuscriptsNo Lean link listed
Number theory#018

The Margulis–Platonov conjecture over global fields

OpenAI source summary

Proves the Margulis–Platonov conjecture over every global field, including function fields of characteristic two. For an absolutely almost simple simply connected algebraic group G over k, every noncentral abstract normal subgroup of G(k) is the inverse image of an open normal subgroup in the finite product of its anisotropic nonarchimedean local groups.

2 manuscriptsNo Lean link listed

The connection to general intelligence

Could better mathematics bring AGI closer?

Progress in proofs can reveal stronger abstraction, sustained reasoning and the ability to build on earlier ideas. Mathematical discoveries could also contribute to better algorithms and scientific tools. If these capabilities transfer reliably across domains, they could accelerate progress towards AGI, potentially in the near future.

That transfer is the open question. A correct theorem does not establish broad autonomy, sound real-world judgment or a reliable arrival date. We track mathematics as one strand of evidence alongside other capabilities.

Follow the AGI Clock ↗

A directory you can check

Claims, manuscripts and formal proofs

One family may contain a main proof, companion results and alternative arguments. A link to Lean scope notes means formal material is available for that family; it does not certify every associated paper. Paper-level listings are shown separately. TaskNorth has not independently checked the proofs.

Corrections remain part of the story

The original release contained 722 manuscripts. The active source catalogue contains 719 after three withdrawals. Withdrawn work is excluded here; the source history records withdrawals, repairs and dependencies.

A dated source snapshot

Checked 2026-10-19. Source links point to the exact repository revision used for this edition. This is a reviewed snapshot, not a live feed. New imports require an editorial check before publication.

What we have simplified

The first 20 guides translate selected families into everyday language. The remaining entries retain attributed source summaries and abstracts, which can be technical. Read the original papers for hypotheses, formulas and full arguments. Source catalogue material is adapted from OpenAI under Apache-2.0.