Claude's Thinking Mode: When a Slower Answer Is Worth It

Most people meet Claude's thinking mode by accident. A progress indicator with a timer appears, the answer takes a little longer than usual, and an expandable section shows up above the response. Some people leave it on for everything on the theory that more thinking must mean better answers. Others never touch it. Both habits cost something, and the useful skill is knowing which tasks repay the wait.

This article separates the two controls people confuse, explains what you actually see, and gives you a quick test for when to use it.

Two controls that sound alike

Anthropic's help page draws a clear line. Effort controls how thorough Claude is on every response. Thinking controls whether Claude works through its reasoning in an expandable section before it answers. They are related, but they are not the same dial.

Where you find them depends on the model:

• On models that have effort levels, click the model name, hover over "Effort," and switch the "Thinking" (or "Extended") toggle on.
• On other models, click the model name and switch the "Extended" toggle on.

One detail changes how you think about the whole feature. According to Anthropic, thinking cannot be turned off on Claude Haiku 5.5, Sonnet 5.5, Opus 5.5, Fable 5.1 or Opus 5. On Opus 5.5 and Fable 5.1 it is always on at every effort level. So on the current models, the choice you actually make is less "thinking or no thinking" and more "how much effort."

What you see while it works

While Claude thinks, a "Thinking" indicator with a timer is shown. Once it answers, an expandable "Thinking" section above the response gives a summary of its reasoning.

That summary is useful, but it is a summary, not a transcript. Read it for what Claude assumed and which approach it chose, not as proof that the answer is right. If the assumptions in the summary are wrong, the answer built on them will be too, and finding that out takes ten seconds.

The cost: time and usage

Higher effort gives more thorough responses, but it takes longer and uses more tokens. Anthropic notes that this means you reach your usage limits faster. If you ask a hundred small questions at maximum effort, you are paying for depth you do not need on any of them.

Anthropic's own list of tasks that benefit is close to the second column: mathematical calculations and proofs, competition-level coding, comprehensive project planning, detailed document analysis and multi-step technical problems. The same page says everyday tasks do not need extra thinking.

A two-question test

Before you raise the effort, ask yourself two things.

1. Would a careful person need to stop and work this out on paper? If yes, extra thinking is likely to help. If the task is something you could answer from memory, it will not.
2. If the first answer is slightly wrong, will I notice? If you will notice immediately, a fast answer and a quick correction is cheaper than waiting. If you will not notice, because the error would hide inside plausible-looking work, thinking time is cheap insurance.

What thinking can't fix

Thinking improves the reasoning Claude does with what it has. It does not give Claude information it lacks.

• If the answer depends on a recent fact, thinking longer will not make it known. Use web search for that.
• If the problem is that your request is vague, more effort produces a more elaborate version of the wrong answer. A clearer request fixes it. The Prompt Framework That Fixes a Vague Claude Answer is the better tool there.
• If the numbers matter, check them yourself. A long chain of careful-looking steps can still contain one wrong figure.

When thinking stops early

Anthropic documents one limitation worth knowing. Thinking may stop before it is complete, with a message saying the rest of the thought process is not available. This happens when the content involves information that safety systems flag as posing a higher risk of harm or misuse. If it affects your response, the help page suggests reframing the prompt. In practice, that means stating the legitimate purpose and context of your question clearly, not trying to word around the filter.

Setting a sensible default

If you use more than one assistant, the same trade-off exists elsewhere: When to Actually Use Think Deeper in Copilot and When Think Mode Is Actually Worth It in Grok cover those tools.

Official sources

• Using extended thinking | Claude Help Center

Checked on October 24, 2026. Features, plans and names change often, so the vendor's own pages are the final word.