Chain-of-Thought

Chain-of-thought is a prompting technique in which a language model is guided to produce intermediate reasoning steps before its final answer, improving performance on multi-step problems.

Generating explicit reasoning helps the model work through tasks like arithmetic, logic, and planning that a single-step answer often gets wrong.

The added reasoning consumes more tokens, raising inference cost, so its benefit on a given task is an empirical trade-off best confirmed by evaluation.