August 3, 2026
Proofs, prices, and panic
Ten advances in mathematics and theoretical computer science
AI says it cracked 10 huge math mysteries, and the comments are already fighting about the bill
TLDR: OpenAI says its upcoming AI helped solve ten long-stuck math problems, then humans turned the work into papers and checked proofs. The comments instantly split between awe and suspicion, with people arguing over whether the headline-grabbing $2,000 cost is real or just the cheapest part of a much bigger experiment.
OpenAI just dropped a very big flex: it says an internal model called Astra produced ten new results in mathematics and theoretical computer science, tackling problems that had reportedly been stuck for years or even decades. The company says the raw model usage to find the solutions would cost about $2,000, then humans helped turn the results into papers and formal proof files. And that one number? Oh, the community immediately pounced.
The loudest reaction was not "wow, math history" so much as "show us the full receipt." In the discussion, skeptics argued the $2,000 figure could be wildly misleading if it only counts the winning attempts and not all the failed runs, researcher time, or engineering effort. One commenter basically compared it to selective stats reporting: if you only reveal the success story, people want to know how many misses came first. Another asked the brutally practical question: how much would it cost to "buy" the same breakthroughs from human experts? That is the kind of comment-section grenade that keeps a thread alive for hours.
But not everyone was doomposting. One popular mood was that this could actually make math feel more mainstream, with human mathematicians becoming the interpreters, explainers, and stars of the next act. And yes, the jokes arrived right on schedule: one commenter laughed at OpenAI saying it would "take responsibility" for proofs that were also formalized in Lean, basically joking that volunteering to vouch for machine-checked math is like offering to be blamed if basic arithmetic goes wrong. In other words: the paper is about proofs, but the comments are about trust, money, credit, and who gets to be the hero now.
Key Points
- •OpenAI announced ten new results on long-open problems in mathematics and theoretical computer science.
- •The company says the results were achieved by an internal version of Astra, its next major model.
- •OpenAI estimates the token cost to find the solutions at roughly $2,000 using Sol API rates.
- •The arguments were prepared into manuscripts by humans with model assistance and then formalized into Lean certificates, which OpenAI released along with reasoning walkthroughs.
- •The article lists results including new bounds in high-dimensional sphere packing and coding theory, a construction proving the existence of non-sofic groups, and a disproof of Connes’s rigidity conjecture.