OpenAI says reasoning model disproved 1946 planar unit distance problem

Details

• OpenAI says one of its models produced a breakthrough proof on the planar unit distance problem, a longstanding combinatorics question first posed by Paul Erdős in 1946.
• The company says mathematicians had long believed the best solutions would resemble square-grid arrangements.
• OpenAI says its model disproved that assumption, marking a notable result on a famous open problem.
• The proof reportedly came from a general-purpose reasoning model, not a system built specifically for mathematics or for this problem.
• OpenAI frames the result as evidence that AI can sustain long chains of reasoning and connect ideas across distant fields.
• The thread links to a longer write-up, suggesting the company is positioning the result as a research milestone rather than a product launch.

Impact

If verified, the result is a meaningful signal for frontier-model competition because it highlights reasoning progress in a domain long used to test AI limits. It also supports the broader shift from chat-style assistance toward models that can contribute to formal research workflows. For OpenAI, the announcement strengthens its case that general-purpose reasoning systems are moving beyond benchmark performance into original problem solving, though claims about mathematical significance will ultimately depend on independent review.