OpenAI just did something that makes every “AI will take your job” headline feel quaint. Its newest model, GPT-5.6 Sol Ultra, generated a machine-verified proof of the Cycle Double Cover Conjecture, a problem mathematicians have been chipping away at for roughly half a century, and it did it in less than an hour.
The proof was published as a PDF on OpenAI’s CDN on July 10, 2026, with authorship attributed entirely to the model itself. Codex assisted with the writeup.
What the conjecture actually says
The Cycle Double Cover Conjecture was posed independently by George Szekeres in 1973 and Paul Seymour in 1979. In English: it claims that for any graph without “bridges” (edges whose removal would disconnect the graph), you can find a collection of cycles that together cover every edge exactly twice.
Partial results existed for specific cases, but a comprehensive, general proof remained out of reach. That is, until an AI system decided to throw 64 subagents at the problem simultaneously.
