01In plain English
A reasoning model spends extra effort thinking through a problem before it replies. It breaks the question into steps, checks its own work and sometimes tries more than one route. This usually makes it slower and more costly to run than a standard chat model.
02How it works
A standard chat model begins writing its answer straight away. A reasoning model first spends a stretch of time working in private. It lays out the problem, breaks it into smaller steps, tries an approach, checks whether the result makes sense and sometimes backs up and tries another. Only after that does it write the reply you see. The extra thinking uses more computing power, so answers arrive later and cost more per question. In exchange, results on problems with several steps, such as a math word problem, a tricky bug or a plan with constraints, tend to be more accurate. It is the difference between blurting the first answer that comes to mind and working it out on scratch paper.
03Why it matters when you are choosing
It tends to do better on math, logic, coding and multi-step planning, where a quick answer is often wrong. For simple writing or lookups it adds delay and cost without much gain.
04What to check
Check whether the tool lets you choose between a fast and a thinking mode, and whether longer reasoning counts against your usage limits. Test it on a hard task from your own work.
05Where you will meet it
06Common questions
When should I use a reasoning model?
Use one for hard problems with several steps, such as math, logic puzzles, debugging code or planning with constraints. For quick tasks like rewriting an email or summarizing a page, a standard model is faster, cheaper and usually just as good.
Why are reasoning models slower?
They spend extra time working through the problem before they answer. That thinking takes computing power and shows up as a delay, sometimes a minute or more on hard questions. Many tools let you choose between a fast mode and a thinking mode.
Do reasoning models still make mistakes?
Yes. Working in steps makes errors less frequent on hard problems, but the model can still reason its way to a wrong answer and sound sure about it. Check important results, especially numbers, legal points and anything you plan to publish.