Chain-of-Thought Prompting

Reviewed by Proof · Updated

Definition

Chain-of-thought prompting means asking a model to work through a problem step by step before it gives the final answer.

Key takeaways

  • Asking for step-by-step reasoning often improves answers on math, logic and multi-step problems.
  • It works on regular chat models. Reasoning Models do something similar on their own.
  • The steps cost output tokens, so answers get longer and pricier.
  • Ask for the final answer in a clear, separate spot so your code can find it.
  • The technique was described in the 2022 paper "Chain-of-Thought Prompting Elicits Reasoning in Large Language Models" by Wei and others.

How to use it

The simplest form is one added line: "Think step by step, then give the answer." You can also show a worked example with the steps written out, which is chain-of-thought combined with Few-Shot Prompting.

To keep the output usable:

  • Ask for the answer on its own final line, or inside tags like <answer>. See XML Prompts.
  • Only show the final answer to users unless they need the working.
  • Leave enough max_tokens for both the steps and the answer.

When to skip it

For simple lookups, rewrites or classification, step-by-step reasoning mostly adds tokens and time. With reasoning models, adding it is often unnecessary, because they already plan internally.

Example

curl https://api.luv13.ai/v1/chat/completions \
  -H "Authorization: Bearer $LUV13_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "luv13/glm-5.3-flash",
    "messages": [
      {"role": "user", "content": "A shirt costs $20 after a 20% discount. What was the original price? Think step by step, then put only the final answer in <answer> tags."}
    ],
    "max_tokens": 400
  }'

The right answer is $25.