Today, OpenAI announced that its next-generation general reasoning model independently produced an original mathematical proof overturning an unsolved geometric conjecture proposed by renowned mathematician Paul Erdős in 1946. "For nearly eight decades, mathematicians believed the optimal solution resembled roughly a square grid structure," OpenAI stated on X. "An OpenAI model has now overturned this assumption, discovering a set of entirely novel constructions that perform better." The company described this as "the first time AI autonomously solved an important open problem within mathematics." However, OpenAI learned lessons regarding trust issues. Seven months ago, former Vice President Kevin Weil claimed that GPT-5 had resolved ten previously unsolved Erdős problems—only for those solutions to be revealed as existing methods already documented in literature, sparking widespread industry mockery. This time, OpenAI was markedly more cautious. Alongside announcing the model's achievement, the company included endorsement statements from multiple mathematicians, including Thomas Bloom, maintainer of the Erdős Problem website, who publicly criticized Weil's earlier claims as "a serious misrepresentation." OpenAI emphasized that this proof came from a general-purpose reasoning model rather than one specifically designed for solving math problems. This indicates that AI can maintain coherence during long-chain, highly complex reasoning tasks while connecting ideas across disciplines—a capability with profound implications for research in biology, physics, engineering, and medicine. "AI is helping us explore more fully the cathedral of mathematics built over centuries," said Bloom. "What unseen wonders await us?"