OpenAI Says Unreleased Model Produced 372 Math and Theoretical CS Results

OpenAI revealed the results in a GitHub repository at 6 p.m. EDT, Slashdot reported. The volume of claims will take mathematicians months to parse and understand, including to determine whether the proofs contain novel and important ideas or are mostly combinations of existing techniques.

Many of the results have already been verified in Lean, a programming language that validates a proof's logic, according to the report. That verification makes them all but certain to be correct, though it does not by itself establish that the work introduces important new mathematics.

Among the results claimed are a solution to the four-dimensional Kakeya conjecture, improvements on some of the world's most important computer algorithms, and actual progress toward the Riemann hypothesis, which the report describes as math's scariest problem.

Scientific American noted that the single-agent approach would be a striking difference from OpenAI's earlier blockbuster solution to the Navier-Stokes problem. That work was produced through the collective efforts of a 10,000-strong agentic swarm and cost millions of dollars in computing power, according to Scientific American.

The GitHub disclosure leaves open questions about novelty and significance. Lean verification can confirm that a proof follows logical rules, but mathematicians will still need to assess whether the 372 results reshape their fields or largely rearrange known methods. OpenAI has not released the internal model described in the disclosure, and mathematicians will need to examine the results outside the company.