OpenAI says its AI model solved a long-standing geometry problem

In short: OpenAI says its new AI model produced an original math proof that overturns a geometry claim that has been open since 1946.

What happened

OpenAI says a new “reasoning model” (an AI system designed to work through problems step by step) created a proof about a famous geometry conjecture first posed by mathematician Paul Erdős in 1946. A conjecture is an educated guess in math that has not been proven true or false yet.

According to OpenAI, the model did not just repeat a known answer. It produced a new proof that disproves the conjecture. The company says the result comes from a general-purpose system, not an AI tool built only for math.

This announcement comes after an earlier embarrassment. About seven months ago, an OpenAI executive said “GPT-5” had solved several unsolved Erdős problems, but the “solutions” turned out to match work that already existed in published math research. This time, OpenAI shared supporting remarks from mathematicians including Noga Alon, Melanie Wood, and Thomas Bloom, who previously criticized that earlier claim.

OpenAI also said mathematicians had long believed the best solutions looked roughly like square grids. The company says its model found a different family of constructions that does better (think of finding a better way to arrange points than the pattern people kept returning to).

Why it matters

If the proof holds up, it would be a concrete example of AI helping experts explore ideas that humans have struggled with for decades. It could also make researchers more willing to test AI on other hard problems, but only when independent experts can check the work carefully.

Source: TechCrunch AI