Chain of thought is the practice of having a model produce its working, the intermediate steps toward a solution, rather than jumping straight to an answer. Spelling out those steps lets the model break a hard problem into smaller ones and gives later tokens more relevant context to condition on.
It started as a prompting trick (“think step by step”) but is now baked directly into reasoning models through training. In that setting the chain of thought is generated automatically and can run to thousands of tokens before the model produces its final response.
