OpenAI claims solved 1946 math problem

OpenAI said on May 20 that one of its internal reasoning models had disproved a longstanding conjecture tied to Paul Erdős’s 1946 planar unit distance problem, a central question in discrete geometry. The company published a proof and companion remarks on its website and said a group of external mathematicians had checked the argument. OpenAI described the result as the first time an AI system had autonomously solved a prominent open problem central to a field of mathematics. Independent peer-reviewed publication had not yet been announced on Thursday. ### Which 1946 problem does OpenAI say it solved? Paul Erdős first posed the planar unit distance problem in 1946: given \(n\) points in a plane, how many pairs can be exactly one unit apart? OpenAI said the problem became one of the best-known questions in combinatorial geometry, and cited the 2005 book *Research Problems in Discrete Geometry* as calling it possibly the best-known and simplest-to-explain problem in the field. (openai.com) OpenAI said the prevailing belief had been that square-grid style constructions were essentially optimal for maximizing unit-distance pairs. The company said its model disproved that conjecture by producing an infinite family of examples that gives a polynomial improvement. ### What exactly did OpenAI claim the model did? (openai.com) OpenAI said the proof came from a “new general-purpose reasoning model,” not from a system trained specifically for mathematics, scaffolded to search proof strategies, or aimed at this problem in particular. The company said it had been testing advanced models on a collection of Erdős problems as part of a broader effort to see whether they could contribute to frontier research. (openai.com) The company also said the proof used ideas from algebraic number theory to address an elementary geometric question. OpenAI framed that cross-field connection as part of why the result mattered for evaluating long chains of reasoning. ### Who outside OpenAI is backing the claim? Fields Medalist Tim Gowers, in the companion paper cited by OpenAI, called the result “a milestone in AI mathematics.” OpenAI also quoted Princeton combinatorialist Noga Alon and number theorist Arul Shankar in its write-up, and TechCrunch reported that companion remarks were published from mathematicians including Alon, Melanie Wood and Thomas Bloom. (openai.com) Firstpost reported that the latest announcement arrived with more cautious backing than an earlier OpenAI claim involving Erdős problems. TechCrunch said Thomas Bloom, who had criticized a previous OpenAI-related claim as “a dramatic misrepresentation,” was among the mathematicians now supporting publication of the new work. (openai.com) ### Why are people treating this claim carefully? Seven months before this announcement, former OpenAI vice president Kevin Weil wrote on X that GPT-5 had found solutions to 10 previously unsolved Erdős problems and made progress on 11 others, according to TechCrunch. Mathematicians later said those results were already in the literature, and Weil deleted the post. (firstpost.com) That episode is one reason the new claim is being read with caution. OpenAI says external mathematicians checked the proof, but broader independent verification, replication and any formal peer-review process remain separate steps from a company blog post and companion remarks. That is an inference from the publication status described in the available reports. (techcrunch.com) ### Why is this landing at a sensitive moment for OpenAI? Reuters reported on May 20 that OpenAI was preparing to confidentially file for a U.S. initial public offering in the coming weeks and was aiming to go public as early as September, according to sources familiar with the matter. CNBC separately reported that a draft prospectus could be filed as soon as Friday and that Goldman Sachs and Morgan Stanley were working with the company. (openai.com) A research claim of this scale is likely to draw more outside scrutiny when a company is also moving toward public markets. Reuters said OpenAI was last valued at $852 billion by private investors. ### What happens next? May 20 is the key date for now: OpenAI has already published the proof and companion remarks. (money.usnews.com) The next concrete step is external checking by mathematicians beyond those named in the company’s materials, followed by any journal submission, referee process or independent write-ups that test whether the argument holds. September is the other date hanging over the story. Reuters reported that OpenAI is aiming to go public as early as then, while the company’s math claim continues to move from company-backed announcement to broader outside review. (money.usnews.com) (openai.com)