The reasoning happens in the chain of thoughts. But OpenAI (aka ClosedAI) doesn't show this part when you use the o1 model, whether through the API or chat. They hide it to prevent distillation. Deepseek, though, has come up with something new.