Reasoning model

Reasoning models are trained to produce an internal chain of thought before their final answer. They work through a problem, try approaches, notice mistakes, and revise, rather than answering in one pass.

This makes them stronger at math, coding, planning, and multi-step analysis. Many current models can switch between quick answers and deeper reasoning, and let you set how much thinking effort to spend.

The tradeoff is cost and latency. Thinking tokens are typically billed, and long reasoning takes time. For simple tasks like classification or formatting, a fast model with little or no reasoning is often the better choice.