OpenAI has rocked the mathematical community by announcing that an internal, general-purpose reasoning model has autonomously disproved a legendary 80-year-old conjecture in discrete geometry: Paul Erdős’s 1946 Planar Unit Distance Problem.

Rather than a minor tweak or a brute-force calculation, the AI generated a novel, cross-disciplinary proof that has been fully checked and verified by some of the world’s leading mathematicians. Fields Medalist Tim Gowers called the achievement “a milestone in AI mathematics” and noted that if a human had submitted it to the prestigious Annals of Mathematics, he would have recommended acceptance without hesitation.

1. The Problem: What Did Erdős Propose?

In 1946, Paul Erdős posed a deceptively simple question: If you place $n$ points on a flat piece of paper, what is the maximum number of pairs that can be exactly one unit apart?

Erdős conjectured that the maximum number of unit-distance pairs, denoted as $v(n)$, grew incredibly slowly—essentially linear, bounded above by near-linear growth:

$$v(n) \le n^{1 + o(1)}$$

For nearly eight decades, mathematicians believed that the optimal layout of points would resemble variations of a square grid (using Gaussian integers).

2. The Breakthrough: How the AI Disproved It

OpenAI’s reasoning model proved Erdős completely wrong by taking a route no human had managed to map out. Instead of tweaking geometric grids, the AI pulled sophisticated tools from algebraic number theory—specifically utilizing infinite class field towers and Golod-Shafarevich theory.

• The Counterexample: The model constructed an infinite family of point configurations showing that the number of unit distances grows at a much faster polynomial rate:

$$v(n) \ge n^{1 + \delta}$$

• The Exact Exponent: In a companion paper analyzing the AI’s math, Princeton professor Will Sawin refined the AI’s construction to pin down a universal constant exponent of $\delta = 0.014$.

This means that as you scale up the number of points into the millions, the AI’s algebraic layouts pack vastly more unit-distance pairs than any grid structure ever could.

3. Autonomy vs. Prior “AI Math” Hype

This breakthrough stands apart from previous AI mathematical milestones due to its sheer autonomy and originality.

In late 2025, tech leaders like Google DeepMind’s Demis Hassabis openly criticized sweeping claims about AI solving Erdős problems, noting that models were often just repackaging known literature found on the web.

This May 2026 result is fundamentally different:

• Zero Human Scaffolding: The model wasn’t given a partial proof to finish, nor was it heavily guided step-by-step by a prompt engineer. It was simply given a benchmark of open problems and worked out the solution entirely on its own.
• Genuine “Cross-Domain” Intuition: Mathematician Thomas Bloom noted that the proof shows an ingenious, unexpected application of number theory to a problem everyone assumed was strictly geometric.

4. Expert Reactions From the Math Community

“This is a really impressive piece of work… I actually briefly worked on this problem and tried to make a counterexample, but failed to make progress.”

— Jacob Tsimerman, Mathematician

“There is no doubt that the solution to the unit-distance problem is a milestone… No previous AI-generated proof has come close to that.”

— Tim Gowers, Fields Medalist