What is Inference-Time Reasoning?

Spending extra computation at answer time — an internal chain of thought — to improve accuracy on hard problems, as used by reasoning models such as OpenAI o1 and DeepSeek R1.