Home / Choosing your model and your tool
Low, medium, high, xhigh, max: choosing your effort level
Start from the level marked Default, go up one notch if the answer seems superficial, go down one notch if it drags on a simple question. The effort level sets how much work Claude puts in before answering: higher effort gives more thorough answers, but they take longer and use up your plan faster.
The effort level sets the amount of work Claude puts in before answering you. According to Anthropic's help page, higher effort gives more thorough but slower answers, and you reach your plan's usage limits faster.
Five levels, from lightest to most thorough
The menu offers Low, Medium, High, Extra high (xhigh) and Max. Low and Medium suit everyday tasks and make your plan last. High offers the best balance between quality and speed. Extra high targets long coding tasks and those where Claude chains many steps on its own. Max, the most consuming, is for problems that call for the deepest reasoning. Extra high only appears on Opus 4.7 and newer models; on an Enterprise account, an administrator may have removed levels from the menu.
Where to set it, and where to start
The rule fits in one sentence: start from the level marked Default, go up one notch if the answer seems superficial, go down one notch if it drags on a simple question.
The path described by the help page takes three clicks:
Model name (next to the send button), then "Effort", then the level you want
The change takes effect from the next answer. The Default level is the one Anthropic recommends for the chosen model, and it changes from one model to another.
Effort and thinking, two separate settings
The same menu carries a Thinking switch: Claude breaks the problem down, plans and tries several leads before answering, and this work appears in a collapsible section with a timer. Effort, for its part, sets how carefully each answer is produced. The two combine freely. On Claude Opus 5.5, thinking cannot be switched off.
In Claude Code, Max lasts one session
In Claude Code, a Max effort chosen in /model applies to the current session only: it is not saved as the default for new sessions. The picker groups older models under a "More models" line. In that picker, the Default label now names your plan's default model, and choosing it in the terminal takes you back to that model. Model choices saved on Fable 5.1 move to the fable alias: right away on Pro and Max plans, and on Team and Enterprise once your organization allows it. To go back, type /model claude-fable-5-1. Ultracode, the setting tied to dynamic workflows, is a separate toggle in /effort (Tab key, or /effort ultracode on): it no longer forces xhigh effort and stays on at any level.
A subagent or teammate running on a different model from the main conversation uses the effort saved for its own model, or that model's default level, and not the session's.
Comparing two levels without fooling yourself
The temptation is to ask a question at High, then at Max, and attribute the whole difference to the setting. But the same question asked twice at the same level already gives two different answers, in wording, length and sometimes duration. So ask the question twice at each level, and only keep an effect if the gap between levels clearly exceeds the gap between two trials at the same level.
Watch the right gauge too. Effort mostly sets the amount of reasoning, which the timer in the Thinking section makes visible. The length of the final text tells you less: the developer documentation states that on Opus 5, changing the effort does not reliably shorten the displayed answer. Choosing a model according to the task rather than by habit rests on the same reflex, and the lesson on moving to Claude Opus 5.5 details effort from the developer side.
The five levels of the Effort menu
| Level shown | For what use | What it costs |
|---|---|---|
| Low | A simple, everyday task, such as rewording a short text. | The most economical, with slightly reduced capability according to the developer documentation. |
| Medium | An everyday task that deserves a little more care. | Economical, it makes the plan last. |
| High | The best balance between quality and speed. | Slower and more consuming than Low and Medium. |
| Extra high (xhigh) | A long coding task, or a task where Claude chains many steps on its own. Opus 4.7 and newer models. | Deeper reasoning than High, without the full cost of Max. |
| Max | A hard problem that calls for the deepest reasoning. | The slowest and most consuming, within the limits of your plan. |
A person sets the effort to High and asks Claude for a follow-up letter to a supplier. The timer in the Thinking section shows 9 seconds, then Claude answers in one paragraph. She opens a new conversation, sets the effort to Max and sends the same request. The timer shows 41 seconds, then Claude answers in three paragraphs. She sent the request once at each level.
Write in one sentence what this situation establishes, and in one sentence what it does not establish.
What this establishes: It establishes that, for this request, the High trial thought for 9 seconds and returned one paragraph, and the Max trial 41 seconds and three paragraphs.
What this does not establish: It does not establish that this gap comes from the change of level, since one trial per level does not distinguish it from the variation between two trials at the same level, nor that the Max answer is the better of the two.
The three most common miscalibrations
- Too broad It establishes that going from High to Max multiplies Claude's thinking time by about four on this kind of letter.
- Too narrow It establishes that the person changed the effort setting between her two sends, without making it possible to compare the two answers obtained.
- Beside the point It establishes that the follow-up letter called for the deepest reasoning Claude can provide.
- Higher effort makes Claude work more before answering, at the price of a longer wait and a plan that runs out faster; in Claude Code, a Max effort applies to the current session only.
- The Default label in the menu marks the recommended starting point for the current model, and that starting point varies by model; in the Claude Code picker, Default names your plan's default model.
- Thinking is a separate switch that gives Claude time to break the problem down, and it cannot be turned off on Opus 5.5.
- A single answer per level does not let you attribute a gap to the setting, because a question asked again at the same level already changes its answer.
- The timer in the Thinking section measures effort better than the length of the text returned.
Pick a question from your everyday work. Leave the effort on the level marked Default, switch on the Thinking switch if it is off, and ask the question twice, in two new conversations, noting each time the duration shown by the timer in the Thinking section. Then set Max and ask it twice more in the same way. Compare the gap between your two trials at the same level with the gap between the two levels, then write in one sentence what these four trials let you conclude.
Every datable claim in this lesson links here to the public text behind it. A source that does not open proves nothing.
- Claude Support, change the model, effort and thinking settings accessed on 2026-09-28
- Anthropic, effort levels for developers, Claude Platform Docs accessed on 2026-09-28
- Claude Code changelog accessed on 2026-10-24
- Claude Code 2.1.317, effort of subagents and teammates on a different model accessed on 2026-11-07
- Claude Code 2.1.319, /model picker: older models under "More models" accessed on 2026-11-07
- Claude Code 2.1.321, saved Fable 5.1 choices switched to fable accessed on 2026-11-13
- Claude Code 2.1.326, /model: the Default label names the plan's default model accessed on 2026-11-17