OpenAI Publishes Solutions to Over 300 Math Problems

On October 6, OpenAI publicly released a batch of mathematical research results generated by its internal frontier models. Its GitHub repository contains 722 manuscripts, categorized into 372 result families, covering main conclusions, supporting arguments, corollaries, and alternative proofs. OpenAI stated that these results address hundreds of open problems but emphasized that their level of verification varies, and some results not formalized in Lean may still contain issues. This release comes about a month after OpenAI claimed to have solved the Navier–Stokes Millennium Prize Problem. However, this conclusion should still be described as a proposed solution by OpenAI, rather than a final result formally recognized by the mathematical community or the Clay Mathematics Institute. OpenAI also made public some Lean proofs, reasoning summaries for 10 results, and computational statistics, stating that it will continue to supplement formal verification. The related materials provide a basis for independent review by the mathematical community, but they are not yet sufficient to prove that all results are correct, nor do they disclose the internal models' specific architecture, training data, or complete reasoning processes. Their academic value still requires long-term peer scrutiny.