Reasoning models internally use a chain-of-thought to break problems down and check solution paths before answering. As a result, they achieve markedly better results on mathematics, logic, and complex code than standard models. The price is higher latency and greater token consumption, since the thinking itself costs tokens. For simple tasks they are often overkill, but for hard tasks they are clearly superior.
Reasoning Model
A reasoning model is an AI model that deliberately thinks before answering and works through intermediate steps to solve complex tasks more reliably. Examples are OpenAI's o-series and Claude with extended thinking.