The Unit Distance Proof

On May 20, 2026, OpenAI announced that one of its internal general-purpose reasoning models had disproved the Erdős unit distance conjecture, an open problem in discrete geometry posed by Paul Erdős in 1946. The question is simple to state: given n points in the plane, what is the largest possible number of pairs lying exactly one unit apart? For nearly eight decades the working belief was that square grid arrangements were essentially optimal.

They are not. The model produced an infinite family of constructions, drawing on Golod-Shafarevich theory and infinite class field towers, that beat the grid by a polynomial factor, achieving on the order of n raised to the power of one plus delta unit distances for some fixed positive delta. The exponent was subsequently sharpened to delta equal to 0.014 by the Princeton mathematician Will Sawin. The argument was checked by a group of external mathematicians who also wrote a companion paper explaining how it works.

The reception among mathematicians was the notable part. Fields Medalist Tim Gowers described the result as a milestone in AI mathematics and said he would have recommended it for acceptance at a top journal without hesitation. Within a week, human researchers had adapted the model's central technique to disprove the sum-product conjecture, a separate open problem. The proof was not merely correct; it was generative.

This was the first time an AI system autonomously resolved a prominent open problem central to a subfield of mathematics. The distinction matters. Machines had assisted proofs for decades, verified them with formal systems, and searched large spaces for counterexamples. What had not happened was a model supplying the core mathematical idea, the choice of an unexpected tool from algebraic number theory applied to a problem in plane geometry, which is precisely the step that mathematicians describe as insight rather than computation.

The follow-on work is what makes it a milestone rather than a curiosity. A result that humans can only verify is a black box; a result whose method humans can absorb and redeploy is a genuine contribution to the field. Mathematicians reading this proof learned a technique and immediately used it elsewhere. That is the ordinary way mathematical knowledge propagates, and it happened here with a machine at the origin. It reframed the question from whether AI could do mathematics to what kinds of mathematics it does well, and how quickly humans can learn from it.