Did ChatGPT just solve an 80-year-old maths problem? OpenAI thinks so

For years, artificial intelligence companies have promised machines capable of scientific discovery rather than simple chatbot conversations. Now, OpenAI says it may have crossed that line with a mathematical result that has stunned researchers.

The company claims one of its new reasoning-focused AI models has independently produced a proof disproving a famous geometry conjecture first proposed by legendary mathematician Paul Erdős in 1946. The problem had remained unresolved for nearly eight decades and was considered one of the notable open questions in combinatorial geometry.

Unlike earlier claims from the company that drew criticism from the academic community, this latest announcement has arrived with cautious backing from several respected mathematicians, suggesting the breakthrough may hold up under scrutiny.

A second attempt after an earlier embarrassment

The announcement inevitably revived memories of an awkward episode earlier this year involving GPT-5.

Seven months ago, former OpenAI vice-president Kevin Weil claimed on X that GPT-5 had solved ten previously unsolved Erdős problems and made progress on several others. The celebration did not last long. Mathematicians quickly pointed out that the AI had merely rediscovered solutions already available in published literature.

More from Tech

Critics including Yann LeCun and Demis Hassabis publicly mocked the announcement, forcing Weil to delete the post.

This time, however, OpenAI appears to have taken a more cautious approach. Alongside its announcement, the company published supporting commentary from mathematicians including Noga Alon, Melanie Wood and Thomas Bloom.

Bloom, notably, had previously criticised OpenAI’s earlier claims as “a dramatic misrepresentation”. His willingness to support the latest development has added credibility to the company’s announcement.

Why the result matters beyond mathematics

According to OpenAI, mathematicians had long believed the best possible solutions to the Erdős problem resembled square-grid structures. The AI model reportedly discovered an entirely different family of constructions that outperformed those assumptions, effectively disproving the conjecture.

More importantly for the AI industry, OpenAI says the proof was generated by a general-purpose reasoning model rather than software built specifically for mathematical research. That distinction matters because it suggests AI systems may now be developing stronger abilities to sustain lengthy chains of logic, test abstract ideas and connect concepts across disciplines.

Researchers believe those capabilities could eventually influence fields far beyond mathematics, including biology, engineering, physics and medicine.

Adding to the X post, OpenAI says, "This result points to something larger: AI systems are becoming capable of holding together long, difficult chains of reasoning, connecting ideas across distant fields, and surfacing paths researchers may not have explored. We believe those same abilities will soon accelerate work in biology, physics, engineering, and medicine."

OpenAI assured that the future still depends on human judgment. "Expertise becomes more valuable, not less. AI can help search, suggest, and verify. People choose the problems that matter, interpret the results, and decide what questions to pursue next," the post says.