Reasoning model

A reasoning model works through a problem step by step in a 'thinking' phase, visible or hidden, before it gives its final answer.

  • Models and architecture
  • Updated

Reasoning models spend extra tokens on intermediate steps, which tends to improve results on maths, coding and other multi-step problems. Some open models include a switch to turn thinking on or off, trading accuracy for speed and cost.

The extra thinking uses context and takes time, so a reasoning model is not always the best choice for simple chat or short answers. When comparing models, check whether a benchmark score was measured with thinking turned on.

Related terms

  • Token A token is a small chunk of text, often a word or part of a word, that a language model reads and writes; models measure length and speed in tokens.
  • Context window The context window is the maximum amount of text, measured in tokens, that a model can take into account at one time, including your prompt and its own answer.
  • AI agent An AI agent is a system in which a model plans and carries out multi-step tasks by calling tools, such as running code, searching the web or editing files.

All terms