Claude's extended thinking is a setting that lets Claude reason through a problem before it replies, with its thought process shown in an expandable section. On older models you switch it on with the Extended or Thinking toggle in the model menu. On Claude Haiku 5.5, Sonnet 5.5, Opus 5.5, Fable 5.1 and Opus 5, thinking can't be turned off; you control depth with the effort level instead.
Anthropic changed how this works with the 5 generation, so a lot of older advice no longer applies. Here's how thinking and effort work today, which models do what and when to change them, from Anthropic's own help centre and developer docs, checked as of October 2026.
Thinking and effort are two different settings
Anthropic separates them clearly. Effort controls how thorough Claude is with every response. The thinking toggle controls whether Claude works through its reasoning in a visible section before replying. You can use any combination of the two, on the models that let you change both.
| Model | Thinking in the Claude app | Default effort |
|---|---|---|
| Fable 5.1 | Always on, every effort level | High |
| Opus 5.5 | Always on, every effort level | Medium |
| Sonnet 5.5 | Can't be turned off | High |
| Haiku 5.5 | Can't be turned off | Medium |
| Older models | Thinking or Extended toggle | Varies |
How to change thinking and effort
- Click the model name next to the send button.
- Pick an effort level. The recommended one for each model is marked Default.
- On models that allow it, switch the Thinking (or Extended) toggle on or off.
- Send your prompt. Changes apply from Claude's next reply, and you can change them part way through a chat.
On an Enterprise plan, an administrator can turn some models or effort levels off for your role, so if an option is missing, ask them first.
The effort levels explained
| Effort | Anthropic's guidance |
|---|---|
| Low and Medium | Work well for routine tasks and stretch your usage further |
| High | The best overall balance of quality and speed |
| Extra high | For long-running coding and agent tasks, deeper than High without the full cost of Max; Opus 4.7 and newer |
| Max | The most thorough option for the deepest reasoning; expect longer replies |
Higher effort means more thorough answers, but they take longer and use more tokens, so you'll hit your usage limits faster. That's the trade-off to keep in mind on a Pro plan.
Which models still show an Extended toggle
Anthropic's help article lists the models with an effort selector: Haiku 5.5, Sonnet 5.5, Opus 5.5, Fable 5.1, Opus 5, Sonnet 5, Fable 5, Opus 4.7, Opus 4.6 and Sonnet 4.6. On those, the toggle may be labelled Thinking or Extended. On older models without effort levels, you'll only see the Extended toggle, and that's the classic extended thinking switch many guides still describe.
So if you're following an older tutorial and can't find the switch, check which model you're on. On the current 5.5 models there's nothing to switch on: thinking is already working, and effort is the setting to adjust.
Reading Claude's thought process
When thinking is on, you'll see a Thinking indicator with a timer while Claude works, then an expandable Thinking section above the reply. Click it to read a summary of how Claude approached the problem. I find this useful for checking a calculation or a recommendation before I rely on it.
Sometimes the thinking stops with a note that the rest isn't available. Anthropic says this happens when its safety systems flag part of the reasoning as a possible risk under its Usage Policy. If it affects your answer, try asking the question from a different angle.
When to raise effort
For everyday writing and simple questions, Anthropic says the defaults work well, and lower effort saves usage. Raise effort, or turn on thinking where you can, for:
- Maths calculations and proofs.
- Hard coding problems.
- Detailed project plans.
- Analysis of long or complex documents.
- Multi-step technical problems.
Whatever you pick, Anthropic's tip is to be specific. A clear prompt helps Claude spend its thinking time on the right thing. Here's one I'd use with high effort:
Copy this prompt
Here's a product list with unit costs and last quarter's sales by product. Work out which three products bring in the most profit after the costs listed, show your working, and flag any number that looks inconsistent.
Work with meWant AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
One practical habit: when you raise effort for a big task, drop it back to Default or Low afterwards. Anthropic says a change applies from Claude's next reply onwards, so if you forget to switch back, every quick question after that runs at Max effort too and quietly eats through your plan.
If your whole team needs to get comfortable with settings like these, see AI workshops for business teams.
A note for developers
On the Claude API, Opus 5.5 and Fable 5.1 always think. Haiku 5.5 and Opus 5 let you turn thinking off at effort levels high and below, but return an error at extra high or max. Sonnet 5.5 returns an error if you set thinking to disabled; Anthropic suggests the between_tools setting, which only thinks between tool calls and is the same speed or faster.
If you want to compare the models themselves, see Claude Opus vs Sonnet, and for longer research jobs, Claude's Research mode goes further than thinking alone.
My verdict
On the newest models you don't need to hunt for an extended thinking switch any more; it's built in. The setting that matters now is effort. Leave it on Default for daily work, push it up for the one task that has to be right, and drop it to Low when you're doing quick admin and want your allowance to last.
For the hardest tasks, see when Claude Fable is worth reaching for.
Frequently asked questions
How do I turn on extended thinking in Claude?
Click the model name next to the send button and switch the Thinking or Extended toggle on, on models that allow it.
Can I turn off thinking in Claude?
Not on Haiku 5.5, Sonnet 5.5, Opus 5.5, Fable 5.1 or Opus 5, according to Anthropic's help centre.
What's the difference between thinking and effort in Claude?
Effort controls how thorough every reply is; thinking controls whether Claude reasons in a visible section first.
Does extended thinking use more of my Claude limit?
Higher effort uses more tokens, so Anthropic says you'll reach your usage limits faster.
Why did Claude's thinking stop early?
Anthropic says this happens when its safety systems flag part of the reasoning as a possible risk.
