Chain of thought
A prompting technique that asks the model to lay out its reasoning step-by-step before committing to an answer — improves accuracy on multi-step tasks.
Chain-of-thought (CoT) is the simplest, most reliable way to make an LLM smarter. Instead of asking for a final answer, you ask for the reasoning out loud, then the answer. The intermediate steps anchor the model and reduce shortcut errors.
In agents, CoT is the bridge between the goal and the next tool call. The model writes out what it knows, what it needs, and what the next step should be before emitting the action.
In 2026, reasoning models (o3, Claude Sonnet 4.6, Gemini 2.5 reasoning) have CoT baked in — they reason internally before responding. For non-reasoning models, prompted CoT remains a routine win.
Frequently asked
Is chain-of-thought still useful with reasoning models?+
Yes, but redundant. Reasoning models do CoT internally. Adding "think step-by-step" to prompts is a no-op or slight negative for them. For non-reasoning models it remains a reliable win.