What is Chain-of-thought?

Prompting (or a model's ability) to reason step by step before answering, which improves performance on complex problems.

Chain-of-thought is the practice of getting a model to work through a problem step by step instead of jumping straight to an answer. Originally this was a prompting trick, such as adding "think step by step" or showing worked examples, and it measurably improved results on math, logic, and multi-step planning tasks. The idea is that each intermediate step becomes context the model conditions on, so errors are easier to avoid and to spot. Modern models often do this automatically, and reasoning models take it further with long internal thinking before the final answer. The tradeoff is more output tokens, which means higher cost and slower responses. It helps most on genuinely complex tasks and adds little on simple lookups or short rewrites.

Example

Asked "A store cuts a $80 jacket by 25%, then adds 10% tax. What is the final price?", a model using chain-of-thought first computes the discount ($60), then the tax ($66), rather than guessing a number in one step.

Why it matters

When comparing AI tools, check whether the model reasons step by step on hard tasks. It often separates models that reliably handle math, code, and planning from ones that only sound fluent. Browse the AI tools directory or the model leaderboard to put it into practice.