OpenAI published mathematical results from an internal frontier model, including Lean formalized proofs and 10 model-reasoning summaries on GitHub.

OpenAI says it is publishing new mathematical results generated by an internal frontier model. The messages report consultation with the Institute for Advanced Study’s independent mathematics-and-AI advisory group, whose advice and public recommendations informed the release approach.

The GitHub materials include substantial Lean proof-formalization code. The reporting describes these formalized proofs as enabling computer checking and says the release also includes paper revisions and citation guidelines.

The disclosure includes 10 model-reasoning summaries and statistics on attempted problems. OpenAI’s reported average compute expenditure per result is approximately equivalent to three hours of ChatGPT Pro deep reasoning; this is a compute-equivalence estimate, not a disclosed GPU-hour count, dollar cost, or elapsed research time.

Further publication and research-support commitments remain plans. OpenAI says it intends to improve paper quality, citations, mathematical presentation, and accessibility to researchers, fund workshops, conferences, and targeted projects, and work toward responsible release of the underlying model; the supplied source set does not include the GitHub repository URL.