"Chain of thought reasoning is built in and always on. The model thinks through problems step-by-step automatically. You do not need to prompt it in any special way. This leads to noticeably better results on anything complex."