OpenAI Model Solves Math Problem That Stumped Humans for 80 Years

An AI model built by OpenAI has done something no human mathematician could accomplish in nearly eight decades. The company announced on May 20 that its general-purpose reasoning model independently disproved the Erdős unit distance conjecture, one of the most famous unsolved problems in discrete geometry. The conjecture was first posed by Hungarian mathematician Paul Erdős back in 1946.

What the Conjecture Said

The unit distance problem asks a deceptively simple question: given a set of points on a flat surface, how many pairs can sit exactly one unit apart? For 80 years, mathematicians believed that square grid arrangements provided the best possible answer to this puzzle. That consensus held firm across generations of researchers and hundreds of papers until OpenAI’s model found a completely different approach to the problem.

Rather than following traditional geometric methods, the model turned to algebraic number theory, a branch of mathematics that explores number systems. It identified hidden patterns inside exotic number fields using advanced concepts like infinite class field towers and Golod-Shafarevich theory. The result was an infinite family of point arrangements that produce significantly more unit-distance pairs than any square grid construction ever could.

Why This Matters for AI Research

What makes this breakthrough especially significant is that the model was not specifically trained for mathematics. It is a general-purpose reasoning system, which means it approached the problem the same way it would handle any other complex task. Fields Medal winner Tim Gowers called the result a milestone in AI mathematics. Princeton mathematician Will Sawin independently verified and refined the proof, expressing the improvement with a fixed exponent.

The discovery suggests that artificial intelligence tools are moving beyond pattern recognition and into genuine scientific reasoning. Number theorist Arul Shankar noted that the work demonstrates AI can generate genuinely original ideas rather than simply recombining existing knowledge.

For the broader research community, this opens a remarkable new chapter in scientific discovery. If a general-purpose model can make frontier contributions to pure mathematics without any specialized training, the potential applications across physics, biology, and engineering could be enormous. OpenAI published the full technical details alongside a companion paper from external mathematicians including Thomas Bloom who independently verified the result.