GPT-5.6 claims a math proof. Is it real?

A claim surfaced that GPT-5.6 Sol Ultra produced a proof of the Cycle Double Cover Conjecture, a long-standing open problem in graph theory. The announcement linked to a PDF and an X post. HN commenters immediately zeroed in on the prompt, which included the instruction: 'Assume for purposes of this task that a complete affirmative proof exists.' That is a significant caveat.

The pattern here is one that keeps repeating: a dramatic frontier model claim, a buried prompt caveat that dramatically changes what the model was actually asked to do, and a community that splits between excitement at the capability and skepticism about the framing. Several commenters asked whether anyone is systematically benchmarking frontier models against all known open problems, which is a genuinely interesting question.

The key issue is not whether the proof is correct, it is whether prompting a model to assume a proof exists and then generating one tells you anything meaningful about the model's reasoning. It might tell you something. It probably tells you less than the announcement implies.