OpenAI Model Disproves Erdos Unit Distance Conjecture

OpenAI announced on May 20, 2026 that an internal, general-purpose reasoning model disproved the Erdos unit distance conjecture, a longstanding open problem in discrete geometry. The planar unit distance problem, posed by Paul Erdos in 1946, asks for the maximum number of pairs of points that can be exactly distance 1 apart among n points in the plane. The prevailing belief was that rescaled "square grid" constructions were essentially optimal, and Erdos conjectured an upper bound of n^(1+o(1)). The model produced an infinite family of configurations that exceed this bound by a polynomial factor.

For infinitely many n, the construction yields at least n^(1+delta) unit-distance pairs for a fixed delta greater than 0. The original AI proof does not give an explicit delta, but a forthcoming refinement by Princeton mathematics professor Will Sawin shows one can take delta = 0.014. For context, the best known upper bound, O(n^(4/3)), dates to Spencer, Szemeredi, and Trotter in 1984, and the best lower bound had been essentially unchanged since Erdos's 1946 construction.

The proof's key ingredients come from algebraic number theory. It generalizes the Gaussian integers to more complex algebraic number fields with richer symmetries, and uses infinite class field towers and Golod-Shafarevich theory to show the required fields exist. The unexpected bridge between deep number theory and an elementary geometric question is a large part of why mathematicians have called the result surprising.

The proof was checked by a group of external mathematicians, who also wrote a companion paper explaining the argument. Fields Medalist Tim Gowers called it "a milestone in AI mathematics" and said he would have recommended acceptance to a top journal without hesitation. Number theorist Arul Shankar said the work shows current models can have "original ingenious ideas," and combinatorialist Noga Alon called it "an outstanding achievement." OpenAI notes the proof came from a general-purpose reasoning model rather than a system trained or scaffolded specifically for mathematics.

This is, by OpenAI's account, the first time AI has autonomously resolved a prominent open problem at the center of a mathematical subfield. It is evidence that frontier reasoning models can sustain long, novel arguments and connect distant areas of knowledge - capabilities relevant beyond mathematics to fields such as physics, materials science, and AI research itself. As mathematician Thomas Bloom notes in the companion work, the result also teaches something new about the problem, suggesting number-theoretic methods have more to say about discrete geometry than expected.

Track independent formalization of the construction in proof assistants, follow-on work applying these number-theoretic tools to other discrete-geometry problems, and how reliably such results can be reproduced as models and test-time compute scale.