Workbranch › Guides

How to use extended thinking in Claude

Claude can stop and reason through a problem before it answers, and you can decide how hard it works. Here's where the switches are, when extra thinking pays off, and what it costs you in usage.

Updated 30 October 2026

Anthropic has changed these controls several times, so older tutorials that mention a "Search and tools" menu may not match your screen. The steps below follow Anthropic's help pages as of October 2026.

Thinking and effort: two separate dials

Claude now gives you two controls, and they do different jobs:

• Thinking (extended thinking) lets Claude take extra time to break a problem down, plan, and try different approaches before it writes the answer.
• Effort sets how much work Claude puts into each reply. The levels are Low, Medium, High, Extra high and Max, and the menu marks each model's recommended level as Default.

The two are independent: you can use any effort level with thinking on or off, on models that let you switch thinking off at all.

Turn thinking on or off

1. Open a chat, new or existing.
2. Click the model name next to the send button. To see older models, click More models.
3. On a model with effort levels, hover over Effort and switch the Thinking toggle (on some models it's labelled Extended).
4. On a model without effort levels, the Extended toggle sits directly in the model menu.
5. Send your message. The new setting applies from Claude's next reply.

On some of the newest models you won't find an off switch. Anthropic lists Haiku 5.5, Sonnet 5.5, Opus 5.5, Fable 5.1 and Opus 5 as models where thinking can't be turned off in the Claude app. On those, use the effort level to control how much reasoning you get.

Choose an effort level

1. Click the model name next to the send button.
2. Click Effort.
3. Pick a level. If you're unsure, start with the one marked Default.

According to Anthropic, the effort menu is available on 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. Extra high only exists on Opus 4.7 and newer.

When extended thinking actually helps

Deeper reasoning is worth the wait when the problem has several steps that depend on each other. Anthropic names these as the strongest cases:

• Maths and proofs
• Competition-level or tricky coding problems
• Planning a project from start to finish
• Detailed analysis of long documents
• Multi-step technical problems

It adds little to small talk, basic facts or a quick email draft. For those, a lower effort level gives you the same answer faster. Whatever the level, a precise prompt helps Claude spend its thinking on the right thing: say what you need, what you already know and what a good answer looks like.

While Claude works you'll see a Thinking label with a timer. When the reply arrives, a collapsible Thinking section above it holds a summary of how Claude approached the problem, which is handy for checking a result before you rely on it.

Good to know: the effect on your limits

• Higher effort costs more. Anthropic says higher effort uses more tokens, so you reach your usage limit sooner. Lower levels make your allowance go further.
• Turn thinking off when you don't need it. Anthropic's usage-limits page suggests switching extended thinking off for tasks that don't need deep reasoning, and choosing a lower effort level for routine work.
• The model matters too. Usage also depends on which model you use, how long and complex the chat is, and which features you turn on.
• Work and school accounts. On Enterprise, an administrator can switch off models or effort levels for your role, or set the default every new chat starts with.

If you keep hitting the cap, see what to do when you reach Claude's usage limit. Long, heavy chats also fill up faster; this guide on long Claude conversations explains how to carry on in a fresh one.

If it doesn't work

• There's no Thinking switch. Your model may have thinking always on (the 5.5 models, Fable 5.1 and Opus 5), or it may be an older model where the switch is labelled Extended directly in the model menu.
• A model or effort level is missing. On an Enterprise plan your admin may have turned it off for your role. Ask them.
• The thinking stops early with a note that the rest isn't available. Anthropic's safety systems cut it off when the reasoning touches content they flag as high-risk. If your request is legitimate, rephrase it so Claude can approach it from another angle.
• Replies are too slow. Drop the effort level a step, or switch thinking off on models that allow it.

Keep the results of a long reasoning session

A deep-thinking answer often ends with a plan full of next steps. With the Workbranch connector for Claude, you can say "save this to Workbranch" and Claude files the chat's tasks, decisions and open questions into a project, so the plan doesn't stay buried in the conversation. See Workbranch for Claude or start a 3-day free trial at app.workbranch.ai.

Sources

• Change the model, effort, and thinking settings (Claude Help Center, checked 30 October 2026)
• How do usage and length limits work? (Claude Help Center, checked 30 October 2026)

Frequently asked questions