According to @godofprompt, an internal OpenAI reasoning model solved an 80-year math problem in one try; nine top mathematicians verified the proof.