[AINews] Quasi-Riemann-Hypothesis: OpenAI publishes 722 math papers solving 90 of the top 500 open math problems; “the most significant moment” in >100 years of mathematics

OpenAI has published a big batch of math work from an internal model, and the reception has been somewhere between amazement and disbelief. The company put 722 manuscripts into a public GitHub repo, grouped into 372 families of related results, along with proof artifacts and selected reasoning summaries. The model itself is still not out in the world. OpenAI says it spoke with the Institute for Advanced Study’s independent Advisory Group on Mathematics and AI before releasing the material.

The headline claim is not that one neat theorem landed. It’s that the collection may contain serious progress across a lot of long-open problems, including work people are already pointing to on Riemann, Hodge and BSD. Commentators singled out the quasi-Riemann result, plus a no-Siegel-zeros result, and one mathematician called it “the most significant moment in mathematical history.” That’s a gigantic statement, and it needs the usual dose of caution. The source itself says these results are being reported by individual commentators and have not been independently verified.

What makes the release hard to ignore is the scale and the weirdly small amount of compute that reportedly produced it. OpenAI says the work came from an evaluation of about 4,000 research problems, and that the average result used roughly three hours of ChatGPT Pro thinking compute. That is not the old image of AI as a brute-force machine chewing through forever-long runs.

There’s also a real split in how people are reading the output. One analysis says about 20% of the results are disproofs or counterexamples, which pushes back on the idea that AI math only wins by searching harder. At the same time, people are already warning that some of the claims won’t survive scrutiny, and others are asking whether this kind of gain in verifiable math and code says anything useful about harder, non-verifiable domains.