loading
loading
Prompting the model to work through its reasoning step by step before giving the final answer, which boosts accuracy on hard problems.
Chain of thought (CoT) means letting the model generate intermediate reasoning — 'first X, therefore Y, so Z' — instead of jumping straight to an answer. It works because the model uses those tokens as scratch space to break a problem into manageable steps, which sharply improves math, logic, and multi-step tasks. The cost is more tokens and latency, and the reasoning shown isn't always the model's true process, so don't trust it as a literal explanation.
Plainly
Think of Chain of Thought as a simple recipe for doing the work better. Prompting the model to work through its reasoning step by step before giving the final answer, which boosts accuracy on hard problems.
In practice
Use it when you need a repeatable method instead of guessing from vibes. In practice, define the owner, input, output, and failure mode before you rely on it.