Reasoning Models

A standard model produces its answer in one pass. A reasoning model is trained to generate an internal chain of thought first — exploring, checking, and revising — before committing to a response, effectively spending more compute at inference time on harder problems. On math, coding, and multi-step logic, that extra “thinking” measurably raises accuracy.

The trade-off is direct: more reasoning means more tokens, higher cost, and slower responses. So reasoning models are worth it for hard, high-stakes tasks and overkill for simple ones — the same right-size-the-model logic that applies everywhere. They bake in what chain-of-thought prompting used to coax out of ordinary models, which is why “just think step by step” matters less with them. For how reasoning models fit alongside prompting and test-time compute, see the AI reasoning overview.