Extended Thinking gives Claude room to think before it speaks. In Claude Desktop (and the Claude app) you turn it on in the chat and, instead of replying immediately, Claude works through the problem step by step first — laying out the logic, checking its own reasoning — and only then gives you the answer. That deliberation pays off on the genuinely hard stuff: multi-step math, tricky logic puzzles, careful planning, anything where a quick gut answer tends to be wrong. It's the chat-side sibling of Claude Code's /effort setting (the Power-Track way to dial up the same thinking).
Ask a hard question and a fast answer is often a confident answer that happens to be wrong — the model blurts the first plausible thing. Extended Thinking fixes that by letting Claude slow down and reason through the problem before it commits to a reply.
You can think of it as the difference between answering off the top of your head and working it out on paper. With the mode on, Claude lays out the steps, follows the logic, catches its own mistakes mid-stream, and only then hands you the conclusion. On problems with real structure — chained arithmetic, constraint puzzles, multi-part plans — that extra care is the difference between right and almost-right.
The trade-off is plain and worth knowing: it’s slower and costs more to run, because thinking longer means doing more work. For “rewrite this sentence” or “what’s a synonym for fast,” that overhead buys you nothing. Reach for it when the problem is genuinely hard and getting it right matters more than getting it now.
If you’ve met Claude Code, this is the chat app’s version of /effort:
the same idea of dialing up how much Claude reasons. For the very hardest problems, turn
on Extended Thinking and pick a more capable model — top brain,
extra thinking time.
why it helps Trade a little speed for a noticeably better answer on problems that have a right answer you can't eyeball.
examples- Save it for hard problems. For a quick question it just makes you wait; the gain shows up on multi-step reasoning.
- It's slower and uses more tokens. That's the trade — more thinking time for a more reliable answer.
- Pair it with the right model. Extended Thinking on a strong model is the combo for the toughest tasks.
Questions people ask
- What does Extended Thinking do?
- It gives Claude room to reason step by step before it answers — laying out the logic and checking its own work first — instead of replying immediately. That deliberation pays off on hard math, tricky logic, and careful planning.
- When should I turn on Extended Thinking?
- Save it for genuinely hard problems with a right answer you can't eyeball — multi-step calculations, constraint puzzles, multi-part plans. For a quick question it just makes you wait.
- Does Extended Thinking cost more or take longer?
- Yes — it's slower and uses more tokens, because thinking longer means doing more work. That's the trade: more thinking time for a more reliable answer on problems where getting it right matters more than getting it now.
- Is Extended Thinking the same as Claude Code's /effort?
- They're the same idea in different places. Extended Thinking is the Claude Desktop / chat version; `/effort` is the Power-Track (terminal) twin — both dial up how much Claude reasons before answering.