TL;DR
OpenAI says its unreleased AI models proved a famous conjecture in every dimension where humans had managed only three, knocked down a basic assumption that hundreds of computer-science papers are built on, and proved part of a Millennium Prize problem in physics. If even one of these holds up under outside review, it could mean AI has started producing original science on its own, the kind of result that has always required a human expert. For now these are company claims. Two of the proofs passed a proof-checking program by OpenAI's account, and no human referee has ruled on any of the three.
Below the line, for paid subscribers: The full story: where it stands, how we got here and what's dated next; our Odds board on 4 milestones, starting with "Outside experts confirm OpenAI's local smoothing proof by 2027"; our labeled speculation on what it adds up to; and every source.