OpenAI Attributes Cycle Double Cover Proof to GPT-5.6 Sol Ultra

OpenAI dropped a PDF on its CDN today attributing a proof of the Cycle Double Cover Conjecture to GPT-5.6 Sol Ultra, the top tier of the model family it launched this week. The story is currently on the Hacker News front page with north of a hundred points and a hundred comments inside its first hour, and the Wikipedia entry has already been edited to note that "On July 10, 2026, OpenAI company claimed the problem was solved using its GPT 5.6 large language model." The claim is fresh enough that it should be read as a claim, not a settled result.

The problem itself is one mathematicians have watched for a long time. The conjecture, as Open Problem Garden states it, is that "for every graph with no bridge, there is a list of cycles so that every edge is contained in exactly two." It was proposed independently by Szekeres and Seymour in the 1970s and is described there as "widely considered to be among the most important open problems in graph theory." A real answer either way would matter.

What is different about this announcement, if it holds up, is the attribution. The credit is aimed at Sol Ultra, a general-purpose language model whose subagent mode OpenAI switched on for the public a day earlier, rather than at a specialised theorem prover or a human-led collaboration where the model played a supporting role. Whether the argument is actually correct is the question graph theorists will spend the next weeks working through, and the honest caveat is that a PDF on a company's CDN is not a peer-reviewed proof, and none of the reporting I could retrieve describes an independent verification.

What the current material does not give you is the internal shape of the argument, the compute involved, how much human editing went into the released text, or which mathematicians OpenAI showed the draft to before posting. Those are the details that decide whether this ends up in the textbook or in the pile of premature announcements. For teams building on these models the direction of travel is the part worth watching, but for now the safer read is to treat the PDF as an opening claim and wait for the graph-theory community to respond.