Reasoning is the extra work a model does to break a problem into steps before it answers or acts. A reasoning setting or reasoning model gives the system more room to work through logic first, which can help on harder tasks and costs more time and usage. In products you’ll see it as a model choice, an effort setting, or a “think longer” mode; the label changes, the tradeoff stays.
Consider an employee sorting mail versus solving a client issue. Sorting mail is fast: look at the name, drop it in the right slot, move on. Solving a client issue takes more thought, comparing options and noticing tradeoffs before deciding. Reasoning is that second kind of work.
Why you care
Reasoning doesn’t guarantee truth. A model can think through bad assumptions and still land on a confident wrong answer, which is why hallucination and review still matter, and why a clear prompt framework helps the model spend its effort on the real problem. Match the thinking budget to the job: fast modes for obvious work like sorting clean transactions, deeper reasoning for designing a workflow or debugging a subtle issue.