August 1, 2026
Proof, Lies & Comment-Section Math
Ten advances in mathematics and theoretical computer science
AI says it cracked 10 old math mysteries — and the comments are already fighting over who gets credit
TLDR: OpenAI says its next AI system found 10 new results in old math problems and that the raw model cost was about $2,000. Commenters are split between amazement and suspicion, arguing over whether this is a breakthrough for science or a carefully marketed number that hides the real cost and credit.
OpenAI just dropped a very big, very nerdy flex: it says an internal AI model found 10 new results in long-stuck math and computer science problems, then humans polished them into papers and checked them in Lean, a proof-checking system. The company even tossed in a jaw-dropper about the computing bill: roughly $2,000 in model usage to find the solutions. And yes, the internet immediately responded with the digital equivalent of "uh, show your work."
That’s where the real drama starts. One camp is surprisingly upbeat, arguing this could drag mathematics out of its ivory tower and into the mainstream. As one commenter basically put it, the sky-is-falling panic from mathematicians feels overblown; people are now actually paying attention, and human experts still matter because they explain why any of this is important. But the skeptics came in hot. The biggest fight? That flashy $2,000 number. Multiple readers are openly suspicious that this sounds a lot like the cheapest winning lottery ticket, not the total cost of all the failed tries, staff salaries, and behind-the-scenes trial and error.
Then came the jokes. One commenter laughed at OpenAI saying it "takes responsibility" for proofs already checked by Lean, comparing it to volunteering to be blamed if basic arithmetic turns out wrong. Another asked the most savage question in the thread: if AI can do this so cheaply, how much would it cost to buy the same breakthroughs from human PhDs? So yes, the paper is about math — but the comments are about trust, credit, money, and whether the humans are being replaced or promoted to hype squad.
Key Points
- •OpenAI announced ten new results on long-open problems in mathematics and theoretical computer science.
- •The company says the results were generated by an internal version of Astra, its next major model.
- •OpenAI estimates the total token cost to find the solutions at about $2,000 using Sol API rates.
- •The arguments were prepared into manuscripts by humans using the same model and then formalized into Lean certificates.
- •The article also discusses attribution and says AI-generated proofs should not be presented as human-authored, while citing concerns raised in the Leiden declaration on AI and Mathematics.