Extended Thinking

What it is

Extended thinking gives Claude more time to break a problem down before answering. With it on, Claude plans, explores different approaches, and reasons through the problem — and you can read a summary of that thought process above the final response.

Best for hard problems that benefit from deliberation: math proofs, competition-level code, multi-step analysis. For a quick factual question, leave it off.

Where to find it

On desktop

Open the model selector in the composer.
Open the Effort submenu and pick a level: Low · Medium · High · Extra · Max. High is the default. Higher levels reason for longer and consume your limits faster.

On the web

Open the model selector in the composer.
Open the Effort submenu and pick a level: Low · Medium · High · Extra · Max. High is the default. Higher levels reason for longer and consume your limits faster.

How to use it

Enable the toggle.
Ask your question — be specific. A vague prompt wastes the extra thinking time.
Watch the Thinking indicator with a timer as Claude processes.
When the response arrives, click the expandable Thinking section above it to read the summary of how Claude got there.

Good to know

Toggling starts a new chat. Turning extended thinking on or off mid-conversation spawns a fresh chat.
Thought process can cut off. If Claude’s reasoning touches something flagged by safety systems, the thinking section will stop short. Reframe the prompt from a different angle if this gets in the way of a real answer.
It takes longer. Responses can be noticeably slower. Only use it when the depth pays off.
Pairs with Research. Enabling Research turns this on automatically — they’re designed to work together.

What’s changed recently

2026-07-29 — The separate Thinking toggle is gone. Effort is now the only control, and it has gained a fifth level: Low · Medium · High · Extra · Max, with High as the default (it was Low). Verified live on claude.ai and against desktop build 1.24012.9.
2026-06-18 — The single Adaptive thinking toggle was replaced by an Effort control (Low / Medium / High / Max, Low is default) with a separate Thinking toggle, both in the model picker’s Effort submenu. Verified on desktop (build 1.13576.4) and web. Tooltip: “Higher effort means more thorough responses, but takes longer and uses your limits faster.”
2026-04-22 — On the web, extended thinking moved from Composer → Search and tools → Extended thinking to the model picker’s Adaptive thinking toggle.
2026-04-22 — On desktop, extended thinking moved from Composer → Search and tools → Extended thinking to the model picker’s Adaptive thinking toggle.
2026-04-18 — Page created from official help-center article.