The o1 series of large language models are trained with reinforcement learning to perform complex reasoning. o1 models think before they answer, producing a long internal chain of thought before responding to the user. o1 is a faster reasoning model particularly good at coding, math, and science.