OpenAI says internal model cracks Erdos unit distance problem
The general-purpose reasoning model produced a polynomial-improvement construction; external mathematicians checked the proof and published a companion paper.
By Ryan Merket ยท
Why it matters
If a general-purpose model can deliver a vetted breakthrough on a classic open problem, founders and operators should expect rapid spillover into code, science, and optimization tasks where correctness can be checked. It is a signal that reasoning is becoming a product surface, not just a demo, and that evaluation frameworks with math-like verifiability will matter for buyers and regulators.
OpenAI said an internal general-purpose reasoning model has produced a proof that disproves a longstanding conjecture in the planar unit distance problem, offering an infinite family of constructions with a polynomial improvement over the widely believed square-grid bounds. The announcement and materials are on OpenAI, with the proof and a companion paper by external mathematicians who checked the argument. In a post on X, OpenAI summarized the announcement as "a general-purpose reasoning model disproves a discrete geometry conjecture."
https://x.com/OpenAI/status/2057176201782075690
The company says the result came from a general model evaluated on a set of Erdos problems, not a system trained specifically for math or scaffolded for this task. The proof applies ideas from algebraic number theory to an elementary geometric question and, according to OpenAI, marks the first time a prominent open problem in a math subfield has been solved autonomously by AI.
In the companion remarks, Fields medalist Tim Gowers calls the result "a milestone in AI mathematics," and number theorist Arul Shankar writes that the paper shows current models "go beyond just helpers to human mathematicians - they are capable of having original ingenious ideas, and then carrying them out to fruition." The remarks situate the work against decades of belief that square-grid constructions were essentially optimal.
OpenAI frames the finding as both a research milestone and an evaluation of emerging reasoning capabilities: precise problems, checkable proofs, and long arguments that only stand if every step holds. The release credits external refereeing and points to further context in the companion paper.