For nearly 80 years, mathematicians believed they understood the limits of one of geometry’s most famous problems. That assumption has now been overturned by an OpenAI reasoning model that independently discovered and provided a proof that a long held mathematical belief was actually false.
External mathematicians have since verified the result, marking what many experts describe as a major milestone for artificial intelligence (AI) in mathematics.
The achievement is drawing attention far beyond the mathematics community because it suggests frontier AI models are beginning to contribute original discoveries in scientific research instead of simply helping humans do basic stuff like write code, summarize documents, or answer questions.
An 80-Year-Old Problem Finally Arrives
The breakthrough centers on the planar unit distance problem, first posed by Hungarian mathematician Paul Erdős in 1946. The problem asks how many pairs of points can be placed exactly one unit apart on a flat plane.
For decades, mathematicians believed the best possible constructions were based on square grid arrangements, and this immediately became a widely accepted conjecture, even though nobody had been able to prove it.
OpenAI says one of its general purpose reasoning models found an entirely new family of mathematical constructions that disproved that assumption. Rather than confirming the accepted idea, the model showed there are infinitely many cases where better arrangements exist.
The company also released the proof alongside reviews from independent mathematicians who examined the work before it was announced publicly.
Mathematicians Call it a Genuine Breakthrough
Several mathematicians who reviewed the work described the result as significant. Princeton mathematician Noga Alon called it an “outstanding achievement,” noting that the solution uses sophisticated tools from algebraic number theory in an elegant and clever way.
Fields Medalist Tim Gowers called the result a “milestone in AI mathematics” and said that if the same solution had been submitted by a human mathematician to the Annals of Mathematics, he would have recommended its acceptance without hesitation.
These reactions are important because the proof was independently examined by mathematicians rather than accepted solely on the AI model’s output.
Why This Matters Beyond Mathematics
The development adds to growing evidence that advanced AI systems are becoming useful for scientific discovery, not just language tasks.
In July 2026, Harvard mathematician Levent Alpöge announced a counterexample to the 1939 Jacobian Conjecture with help from Anthropic’s Claude Fable 5. While these two developments came roughly two months apart and involved different mathematical problems, both show frontier AI models contributing to research questions that had remained unresolved for decades.
The OpenAI result, therefore, sits within a rapidly developing area of AI research. Human mathematicians still also have to examine and verify such work, but these cases show that frontier models are beginning to produce mathematical results that experts consider genuinely new.
