Self-Consistency
Reasoning
Level 3 · Agentic & compositeGenerate several independent reasoning paths for the same problem and take the most common answer.
Self-consistency extends chain-of-thought reasoning by generating multiple independent attempts at solving the same problem — rather than trusting a single reasoning path — and then selecting the answer that appears most frequently across them. Because a model's occasional reasoning slips tend to be inconsistent from one attempt to another, while correct reasoning tends to converge on the same answer, taking the majority result measurably improves accuracy on tasks with one objectively correct answer.
When to use it
Use it for problems with a single objectively correct answer — math, logic, multi-step calculations, classification — where a wrong turn early in the reasoning can silently produce a confident-sounding wrong answer.
When to avoid it
Avoid it for open-ended or creative tasks with no single correct answer (there's no meaningful 'majority' to compute), and for simple tasks where one chain-of-thought attempt is already reliable enough that generating several just adds cost.
Advantages
- Measurably reduces reasoning errors compared to a single chain-of-thought attempt
- Catches inconsistent reasoning slips that a single attempt wouldn't reveal
- Requires no change to the underlying task — only to how many times you ask and how you combine the answers
Disadvantages
- Multiplies token cost and latency by the number of attempts generated
- Only meaningful for tasks where 'the most common answer' is a real signal — i.e. a single correct answer exists
- Doesn't fix a systematic error that appears the same way in every attempt
Examples
A store had 120 apples. It sold 35% of them in the morning and 40 more in the afternoon. How many apples are left? Generate 5 independent step-by-step solutions to this problem, reasoning separately each time. Then compare their final answers and report the one that appears most frequently, along with how many of the 5 attempts agreed.
Pairs well with
Try it
Fill in the placeholders below and see the compiled prompt update live.
What you want to achieve.
The material to work from.
How many independent solutions to generate. Default: 5.
- 34
- Words
- 262
- Characters
- 0/3
- Placeholders filled
- 1 min
- Reading time
- Completeness0
- Clarity0
- Context0
- Specificity0
- Constraints0
- Format0
- Output0
- Constraints note0
- Success criteria0
- Scope focus100
- Verification step0
- Negative example0
- Confidence0
- Placeholder syntax100
- Markdown structure0
- Fill in the remaining placeholders.
- Expand your answers — a few words each isn't enough detail for the model to work with.
- Add background context so the model understands the situation.