Reasoning models are AI models trained or prompted to spend additional computation decomposing, checking, and revising a problem before returning an answer. This guide explains the mechanism, trade-offs, evaluation, and controls that matter in practice.