OpenAI Releases 722 Math Papers
OpenAI publishes internal model's solutions to hundreds of open math problems with compute costs and Lean proofs.
ImportanceStructuralEvidenceE3 inspectableWrite-upDeep
OpenAI released 722 mathematical manuscripts on GitHub on October 6, claiming its internal frontier model solved hundreds of long-standing open problems.
In September, OpenAI only announced solving "more than 100" open problems without details. This release responds to recommendations from the newly formed independent advisory group AGMAI, aiming to alleviate academic concerns about AI racing to publish results.
Each paper used approximately three hours of ChatGPT Pro compute on average. The repository includes reasoning summaries, compute estimates, and Lean formalization code for community verification. OpenAI committed to improving citation and presentation quality in future releases.
The results have not yet undergone full peer review, and the mathematical community is assessing their correctness and ethical implications.