OpenAI Claims AI Solved 50-Year-Old Math Problem in Under an Hour

OpenAI asserts that its artificial intelligence has solved a complex mathematical problem that has stumped human mathematicians for over half a century. The company claims its GPT-5.6 Sol Ultra model generated a proof for the 'cycle double cover conjecture' in less than an hour, utilizing 64 sub-agents.

According to an X post by OpenAI researcher Sol Knight, the AI model was directed to assume a complete positive proof exists and to avoid partial results or reductions to other unsolved conjectures. The 'cycle double cover conjecture,' proposed in 1973 by Hungarian mathematician György Szekeres, posits that every bridgeless undirected graph can be decomposed into a set of cycles that cover each edge exactly twice.

Knight has released a two-page prompt and a three-page final proof document to support the claim. However, for OpenAI's achievement to be officially recognized, the proof must undergo rigorous independent verification by mathematicians and a formal academic review process. The conjecture has seen previous claims of solutions that were not accepted by the mathematical community.

If validated, this would mark a significant milestone, with a general-purpose AI independently discovering a complete proof for a major unsolved mathematical problem. This development could signal a paradigm shift in AI-driven scientific research, showcasing the potential of multiple AI agents exploring diverse problem-solving strategies.

“

GPT-5.6 Sol Ultra, using 64 sub-agents, generated a proof for the 'cycle double cover conjecture' in less than an hour.