Chain-of-Thought (CoT)
Chain-of-thought prompting has a model work through intermediate reasoning steps before answering — improving accuracy on multi-step problems.
Chain-of-thought prompting has a model write intermediate reasoning steps before its answer. It works because each token it writes becomes context for the next, giving the model scratch space. Its 2026 status is nuanced: reasoning models generate thinking tokens natively, so the instruction is often redundant cost, though it still helps on weaker tiers.
Chain-of-thought (CoT) is the technique of having a language model produce intermediate reasoning steps before its final answer — decomposing a problem in writing instead of jumping to a conclusion.
It works because generation is sequential: each reasoning token the model writes becomes context for the next, effectively giving the model scratch space. On arithmetic, logic, and multi-step planning, eliciting steps ("think step by step", or few-shot examples that demonstrate worked reasoning) historically delivered large accuracy gains.
Its 2026 status is nuanced: CoT prompting became less necessary as reasoning models internalized the behavior — they generate thinking tokens natively, and redundant "think step by step" instructions can just add cost. The technique still matters on non-reasoning tiers, in LLM-as-judge rubrics where visible reasoning aids auditability, and as the conceptual ancestor of both branching methods like Tree of Thoughts and the reasoning-model era. When to reach for explicit CoT versus structure versus examples is mapped in Few-Shot vs Chain-of-Thought vs Structured Prompting.
Frequently asked questions
- Does 'think step by step' still help on modern models?
- Less than it used to, and sometimes not at all. Reasoning models think internally by default, so an explicit CoT instruction is often redundant — and can add latency without accuracy. CoT prompting still earns its keep on non-reasoning models and on tasks where you want the steps visible for auditing.
- Is chain-of-thought the same as a reasoning model's thinking?
- Same idea, different layer. CoT is a prompting technique — you ask the model to externalize steps in its answer. Reasoning models bake it in: they're trained to generate internal thinking tokens before responding, no prompt trick required. The technique became the architecture.
Filed under
chain-of-thought · prompting · reasoning · llm
Related
- Few-Shot PromptingFew-shot prompting includes worked examples in the prompt so the model learns the task's pattern from demonstrations instead of instructions alone.
- Reasoning ModelA reasoning model is an LLM trained to think before answering — generating internal reasoning tokens it can spend adaptively on hard problems.
- Few-Shot vs Chain-of-Thought vs Structured Prompting: What to Use When (2026)When to reach for few-shot examples, chain-of-thought reasoning, or structured/output-constrained prompting — a 2026 decision guide to the core techniques.
- Prompt Patterns for Coding AgentsPractical prompting patterns: chaining, few-shot, context management, tool use, and output structuring.
- Extended ThinkingExtended thinking is the reasoning tokens a model generates before its final answer, trading latency and cost for higher accuracy on hard problems.
- Tree of ThoughtsTree of Thoughts is a prompting method that explores multiple reasoning branches as a search tree, evaluating and backtracking among them.
- Prompt EngineeringPrompt engineering is the practice of designing an LLM's inputs — instructions, context, examples, and format — to reliably get the output you want.
- ReAct (Reasoning + Acting)ReAct is an agent loop that interleaves reasoning with tool actions — Thought, Action, Observation, repeat — so the model plans, calls a tool, and revises.