“A version of the O1 model scored higher on the MATH benchmark than any of OpenAI's previous bespoke systems designed for math problems.”