OpenAI is making headlines again, this time claiming its new reasoning model has cracked a famous unsolved problem in geometry. The conjecture, posed by mathematician Paul Erdős in 1946, has stumped researchers for nearly 80 years. According to OpenAI, the AI found an entirely new family of mathematical constructions that outperforms what mathematicians believed was the best possible solution.
It’s a stunning claim. It’s also a claim that lands OpenAI in a familiar position: making bold announcements about AI solving hard problems.
The Ghost of Kevin Weil’s Post
Seven months ago, then-VP Kevin Weil posted on X that GPT-5 had solved 10 previously unsolved Erdős problems and made progress on 11 others. The post made waves. It was precisely the kind of headline that feeds the AI hype cycle: a cutting-edge model cracking decades-old mathematical puzzles.
Then reality caught up. It turned out GPT-5 hadn’t solved anything new. The model had simply rediscovered existing solutions already buried in the mathematical literature. Rivals like Yann LeCun and Google DeepMind CEO Demis Hassabis didn’t hold back their criticism. Weil deleted the post. It was a public stumble that exposed the gap between marketing and substance at a major AI lab.
That context matters enormously for understanding this week’s announcement.
This Time With Backup
To its credit, OpenAI seems to have learned something from the backlash. The company didn’t just drop a claim into the void this time. Alongside the announcement, OpenAI published supporting statements from respected mathematicians including Noga Alon, Melanie Wood, and Thomas Bloom, who maintains the Erdos Problems website.
Bloom’s validation carries weight, especially given that he previously called Weil’s post “a dramatic misrepresentation.” His willingness to endorse this new claim suggests the company has been more careful about vetting before speaking publicly.
OpenAI framed the discovery as a milestone for AI reasoning broadly. The company said this marks “the first time AI has autonomously solved a prominent open problem central to a field of mathematics” and emphasized that the model used wasn’t specifically built for mathematics or this problem. According to OpenAI, the proof emerged from long chains of reasoning and cross-disciplinary idea-connecting that humans may not have previously explored.
If true, that’s significant beyond the specific geometry problem. It would demonstrate that general-purpose AI systems can hold together complex reasoning across domains in ways that might have applications in business, biology, physics, engineering, and medicine.
But Wait
Even with mathematician co-signers, healthy skepticism is warranted. The question isn’t whether OpenAI has fabricated something entirely, but whether the framing obscures what actually happened. Did the AI independently discover a proof that human mathematicians had somehow overlooked? Or did it synthesize existing ideas in a novel way? Those are different achievements with different implications.
Bloom’s public statement suggests genuine mathematical value here. “AI is helping us to more fully explore the cathedral of mathematics we have built over the centuries,” he said. “What other unseen wonders are waiting in the wings?”
But that’s cautious language. It’s exploration and discovery, not necessarily groundbreaking innovation. There’s room between genuine mathematical contribution and world-changing breakthrough.
The Larger Pattern
What’s worth noting is that OpenAI keeps pushing the boundary on what it claims its systems can do. Sometimes those claims hold up. Sometimes they collapse spectacularly. The company has every incentive to announce major accomplishments, and limited downside if the claims turn out to be overstated in hindsight.
The math community’s role matters here. When researchers like Bloom engage publicly and stake their reputations on a claim, it adds a layer of accountability that pure company announcements lack. That’s how you separate genuine breakthroughs from marketing theater.
For now, the proof exists. Mathematicians are examining it. The real test isn’t OpenAI’s press release, but whether the mathematical community ultimately accepts this as a meaningful contribution to the field. That might take time. And in an era where hype moves faster than verification, that gap between announcement and acceptance is where the real story lives.
