Reasoning model
A model trained to produce a long internal working-out before its answer. More accurate on hard problems, slower and dearer.
Also written: reasoning models, thinking model, extended thinking.
A reasoning model is trained to generate a long chain of intermediate steps before committing to an answer. On problems with a right answer that needs several steps — maths, debugging, planning — this helps substantially.
The catch is that those intermediate tokens are output tokens, and you are billed for them at the output rate even though most interfaces do not show them to you. A reasoning model can quietly cost five or ten times a conventional one on the same question, and it is slower.
Use one where being wrong is expensive. For drafting an email, it is an expensive way to be slower.