Step-by-Step Reasoning (Chain of Thought)

Space to work it out

For multi-step problems (arithmetic, logic, planning, code analysis) accuracy often improves when the model writes out intermediate steps before the answer, because each generated step becomes context for the next one. Typical phrasing: "Work through this step by step, then give the final answer on a line starting with Final answer:". Putting the answer on a fixed line makes it easy to parse. Note that newer reasoning models do this internally, and for them forcing long visible steps may add nothing; check your model's guidance. Never treat the explanation as proof: a fluent chain can still contain a wrong step.

Think in steps, check the result

Complex tasks improve when broken into steps, sampled several times or verified.

Parsing a step-by-step reply, run

I ran this plain-Python (standard library only) example. The reply (written by hand to illustrate the format) shows steps and a Final answer: line. The code extracts 54 and independently recomputes 12 x 5 minus 10% = 54.0.

import re

reply = """Step 1: 12 pens at 5 rupees each cost 12 * 5 = 60.
Step 2: A 10 percent discount is 6.
Step 3: 60 - 6 = 54.
Final answer: 54"""

m = re.search(r"Final answer:\s*(\d+)", reply)
print("parsed answer:", int(m.group(1)))
print("check:", 12 * 5 - 0.10 * (12 * 5))

Output:

parsed answer: 54
check: 54.0

Verify with code when you can

For arithmetic and logic, recompute the answer in code instead of trusting the written steps.

Quick check: Why put the final answer on a fixed line?

• So your code can parse it reliably
• To hide the reasoning
• To lower the price
• To skip the steps