OpenAI says a new general-purpose reasoning model has produced an original proof that disproves a geometry conjecture unsolved since 1946, a problem originally posed by mathematician Paul Erdős. TechCrunch reports that the company frames the result as significant because it comes from a reasoning-focused model rather than a system designed specifically for mathematics. OpenAI also says the advance reflects improved ability to follow long, difficult chains of reasoning and connect ideas across different fields, with potential implications for areas such as biology, physics, engineering, and medicine.
Slashdot notes that this is not OpenAI’s first public claim about solving major longstanding math problems. It points to a prior, earlier episode where a former OpenAI executive posted on social media about “GPT-5” finding solutions to multiple unsolved Erdős problems, which critics later said were already known. In the current case, OpenAI reportedly publishes supporting “companion remarks” (a PDF) alongside its announcement, and mathematicians including Noga Alon, Melanie Wood, and Thomas Bloom are described as backing the disproof or reviewing it. Slashdot also contrasts the present claim with the earlier one that was called a misrepresentation by the same site and researchers.
