OpenAI publishes new math breakthroughs and computer-verified proofs

OpenAI just released a suite of new math results from an internal model, including preprints claiming integer multiplication below traditional bounds.

The company published the findings in a public GitHub repository following recommendations from an advisory group at the Institute for Advanced Study. To help researchers verify the math, OpenAI included computer-checked proof formalizations written in Lean, along with reasoning summaries and attempt statistics. Generating the average result took the compute equivalent of about three hours of ChatGPT Pro thinking time.

Why it matters: AI models can sound confident while hallucinating wrong answers. Publishing computer-checkable Lean proofs lets mathematicians verify the work directly while showing the exact compute costs behind the reasoning leaps.

Know this: OpenAI says it is working on a responsible release of the model itself and plans to fund upcoming workshops to discuss the findings with the scientific community.

Now comes the fun part: seeing if the proofs hold up under real academic scrutiny.