OpenAI publishes 722 mathematical manuscripts produced by an internal model

OpenAI says an unreleased internal model produced the 722 manuscripts after being evaluated on approximately 4,000 problems. It estimates that each result used an average of three hours of ChatGPT Pro thinking compute. The repository includes Lean formalizations for many manuscripts and abridged reasoning summaries for selected results. OpenAI cautions that some unformalized results could have issues.