IIRC, GPT-o1 was a reasoning model, but the reasoning traces weren’t exposed to the user until after DeepSeek R1 did so.