Reasoning model
A model that works through a problem in steps before answering instead of replying immediately. It does better on logic, math, planning, and code than a standard model, and worse on nothing except speed and price. In chat apps it shows up as a "thinking" mode that takes longer and uses more tokens.
In plain English
A regular model answers like someone talking off the top of their head. A reasoning model scribbles on a napkin first. Same brain, more napkin. It helps on hard problems and wastes time on easy ones.1
In the real world
- The “think longer” option in ChatGPT.
- Claude’s thinking mode, where the model decides how much to think per request.2
- The spinner that says “thinking…” for thirty seconds.
When to use it
- Planning with several moving parts.
- Math, spreadsheets, logic puzzles, code that has to actually run.
- Any question where being wrong costs more than waiting.
When not to use it
Quick rewrites, summaries, “make this friendlier.” You’ll wait longer for the same answer and pay more for it.
Good, bad, ugly
- Good: fewer dumb mistakes on hard problems.
- Bad: slower and more expensive on everything.
- Ugly: the visible “thinking” looks like insight into the model’s mind. It’s more text, generated the same way.
Key facts
- In the app
- "Think longer," "extended thinking," or a thinking toggle
- Costs
- Slower replies and more tokens
- Best for
- Math, logic, multi-step planning, debugging
Common questions
Should I always use the thinking mode?
No. For a quick rewrite it's slower for no gain. Turn it on for problems with several steps or where a wrong answer is expensive.
Is it actually reasoning?
It generates intermediate steps before the final answer, and that measurably improves results on multi-step tasks. Call it what you like; it works.
Why is it more expensive?
The thinking is text too. Every step it writes to itself counts toward your usage.
References
- Reasoning modelWikipediaView source
- Extended thinkingAnthropic, Claude Platform DocsView source