Chain of Thought

Chain-of-thought prompting elicits intermediate reasoning steps, which raises accuracy on math, logic, and multi-hop questions. Reasoning models internalize this with test-time compute — spending more tokens “thinking” before the final answer.