The short answer
Claude Code’s model-configuration documentation confirms extended thinking and explains how it relates to effort. It defines extended thinking as the reasoning Claude emits before responding. On models that support adaptive reasoning, effort is the primary control for how much thinking happens. So /effort is relevant to thinking depth—but it is not itself a binary extended-thinking toggle. (Claude Code model configuration.)
The same documentation also refers to settings that turn thinking on or off and control how it is displayed. Their availability and behavior can depend on the model; the documentation does not establish one universal default or toggle instruction for every model. Keep this separate from Claude chat’s “Thinking” or “Extended” toggle, which is a different product surface. (Claude Code model configuration; Claude Help Center settings article.)
How to adjust effort in Claude Code
In a Claude Code session:
- Use
/modelto select a model. - Type
/effortto open the documented effort selector, then choose an available level. - Give Claude the task and review its changes and test results as usual.
Claude Code documents /effort as a way to adjust reasoning effort; its release notes describe the no-argument command as an interactive slider. Available levels depend on the selected model. Use the selector rather than assuming a particular level is available. (Claude Code command reference; Claude Code Week 16 release notes.)
Effort is broader than thinking time: Anthropic says it also affects how much work Claude does overall, including how many files it reads, tools it uses, and steps it takes before checking back in. (Choosing a Claude model and effort level in Claude Code.)
When might higher effort be useful for coding?
| Workload | Practical starting point | What the guidance supports |
|---|---|---|
| Small, routine edit or straightforward question | Leave the model’s default effort | Anthropic says defaults work well for everyday tasks; lower effort can stretch usage further. This is general Claude settings guidance, not a Claude Code-specific performance result. |
| Complex coding or agentic work | Consider a higher level if the selected model offers it | Anthropic’s model-settings guidance recommends trying xhigh for complex coding and agentic tasks on Opus 4.7 or newer. This does not establish that every Claude Code model or provider exposes xhigh. |
| Difficult, correctness-critical work | Consider the highest level available, then verify the result | Anthropic describes Max as an option for the most difficult correctness-critical work and says to expect longer response times. More effort is not a guarantee of correctness. |
The general trade-off Anthropic describes is that higher effort can produce more thorough responses, but takes longer, uses more tokens, and can use up usage limits faster. No particular token total, price, or coding-quality improvement follows from selecting a higher level. (Claude Help Center settings article.)
How this differs from the production Claude API
Claude Code’s /effort command is a session control. If you are separately building an application with the Messages API, follow that API’s model-specific documentation instead; do not assume Claude Code’s controls and API request settings are interchangeable.
For Claude Opus 5.5 API requests, Anthropic’s migration guidance says adaptive thinking is always on. It says requests can omit the thinking field or specify adaptive thinking, and that disabling thinking or sending a manual budget_tokens thinking budget is rejected. The same guidance identifies effort as the request parameter that controls thinking depth. These are Opus 5.5 API statements—not instructions for Claude Code, and not a complete request example. (Claude Opus 5.5 migration guide.)
Bottom line: Claude Code supports extended thinking; for models with adaptive reasoning, effort is the primary control over how much thinking happens. Use /effort to adjust effort, but don’t treat it as a universal on/off switch. For a production API integration, use the current request guidance for the exact model you call.
Sources
Model configuration - Claude Code Docs
Week 16 · April 13–17, 2026 - Claude Code Docs
All settings - Claude Code Docs
Choosing a Claude model and effort level in Claude Code
Change the model, effort, and thinking settings - Claude Help Center