OpenAI has published full or partial solutions to 372 previously unsolved mathematics problems. The release, dubbed the mathpocalypse, has stunned the academic world.

Earlier this week, the company posted 722 manuscripts to GitHub. It says an unreleased internal AI model produced them, averaging about three hours of compute per solution. The papers span algebra, number theory, computer science, logic and topology.

OpenAI claims progress on the Riemann hypothesis and the Kakeya conjecture. Three papers were retracted within a day. Some proofs were certified by the automated checker Lean, but many remain unverified.

Reactions are split. Christopher Hacon called it a watershed moment. Scott Aaronson compared it to a resort appearing beside a rainforest survivor. Andrew Sutherland urged treating single-agent claims as unverified until the model is released.

The Association for Human Mathematics urged mathematicians to stop working with OpenAI. Terence Tao shared the statement. Verification of the dense, hundred-page proofs could take months.

The episode raises a deeper question: what counts as mathematical understanding when machines produce proofs humans cannot yet check? The answer will shape the field for years.