Extended Thinking

What it is

Documentation update — 7 October 2026. The model guide now includes Sonnet 5.5 and Opus 5.5. Thinking is mandatory in chat for both, Fable 5.1 and Opus 5. Recommended effort is model-specific; Enterprise administrators may restrict choices or set starting defaults. The effort submenu was inspected on both platforms with Opus 5.5 selected on 7 October.

Effort controls how much thinking Claude puts into a response. Turn it up and answers get more thorough, but they take longer and use more of your limit; turn it down and routine work stretches your usage further.

It lives inside the model picker, not on the composer itself — the pill’s label shows the current level alongside the model.

Where to find it

On desktop

Click the model pill on the right of the composer.
Hover Effort. The submenu opens with five levels — Low, Medium (marked Recommended), High, Extra and Max — above the line “Higher effort means more thorough responses, but takes longer and uses your limits faster.”

On the web

Same two steps, same five levels and the same line of help text on claude.ai.

How to use it

Open the model pill and hover Effort.
Pick a level:
- Low and Medium suit routine work and stretch your usage further.
- High gives the best all-round balance of quality and speed.
- Extra is built for long-running coding and agentic tasks.
- Max is the most thorough option, for work that needs the deepest reasoning. Its row spells out a usage warning in the label itself.
Send your question. Be specific — a vague prompt wastes the extra thinking.

Good to know

Effort and the model are separate settings, and you can combine them however you like. The effort selector works with 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.
No Thinking toggle was visible with Opus 5.5 selected on 7 October. Anthropic’s article still describes one, and also says thinking is mandatory on Fable 5.1 and Opus 5 in the chat interface — which would explain its absence with those models selected. If you’re on an older model and see the toggle, that’s the one the article means.
Default effort depends on the model. Medium was marked Recommended for the inspected Opus 5.5 selection; the current guide does not promise that for every model.
The level you have selected is your account’s setting, not a product default.
Higher effort is slower. Only reach for it when the depth pays for the wait.

What’s changed recently

2026-10-07 — The inspected Opus 5.5 effort menu labels Medium as Recommended and Max as 4× or more usage on desktop and web. Other model defaults were not inspected.
2026-09-13 — The Default marker sits on Medium, where this page previously said High. The five level names — Low, Medium, High, Extra, Max — are the same on desktop and web and in both Chat and Cowork mode. Verified on desktop build 1.52386.3.
2026-09-13 — Removed the “toggling starts a new chat” and “expandable Thinking section with a timer” notes: neither is in Anthropic’s current article and neither can be checked from the interface.
2026-07-29 — The separate Thinking toggle disappeared from the Effort submenu, and Effort gained a fifth level, Extra.
2026-06-18 — The single Adaptive thinking toggle was replaced by an Effort control plus a separate Thinking toggle, both inside the model picker.