Claude Code /effort Explained: Control How Hard Your AI Thinks

Why You Need /effort

When developing with Claude Code, you’ve probably noticed something:

Sometimes you just want Claude to rename a variable, but it spends 10 seconds “deep thinking” and writes an entire analysis. Other times you want it to carefully debug a complex issue, but it fires back a shallow answer in a flash.

The problem? Claude’s “thinking intensity” doesn’t match your task complexity.

Simple tasks get over-analyzed, wasting time and tokens. Complex tasks don’t get enough reasoning depth, hurting quality.

That’s where /effort comes in — a switch that lets you control how deeply Claude thinks.

What Is /effort

/effort is Claude Code’s reasoning effort adjustment command. It controls how much “thinking resources” Claude invests when answering questions, directly affecting response speed, token consumption, and output quality.

In interactive mode, type:

/effort medium

The status bar will display the current effort level icon, confirming the setting is active.

Four Levels

/effort offers four levels, from lightweight to full power:

low (Lightweight)

/effort low

Quick, direct implementation with minimal overhead.

• Status icon: ○ (empty circle)
• Best for: Simple renames, formatting, one-line code changes, quick Q&A with known answers
• Characteristics: Fastest responses, lowest token consumption, but not suited for tasks requiring deep analysis

medium (Balanced)

/effort medium

Balanced implementation with standard testing and verification.

• Status icon: ◐ (half-filled circle)
• Best for: Most everyday development tasks — writing functions, fixing bugs, adding features
• Characteristics: The sweet spot between speed and quality, also the recommended default for Opus 4.6
• Bonus: Using medium maximizes your rate limit quota

high (Deep)

/effort high

Comprehensive implementation with thorough testing and documentation.

• Status icon: ● (filled circle)
• Best for: Complex refactoring, multi-file changes, bugs requiring careful analysis, architecture design
• Characteristics: Claude spends more time thinking, producing more detailed and complete output

max (Full Power)

/effort max

Maximum capability, deepest reasoning.

• Status icon: ◉ (circle within circle)
• Best for: The most complex tasks — large-scale refactoring, cross-system design, scenarios demanding peak reasoning
• Limitation: Only supported on Opus 4.6. Other models automatically downgrade to high
• Note: Significantly higher token consumption and noticeably longer response times

Special Mode: auto

/effort auto

Resets to the model’s default effort level. Defaults vary by model and subscription type:

• Opus 4.6 (Pro/Max/Team subscriptions): Defaults to medium
• Other cases: Defaults to high

auto essentially means “stop manual tuning, let the system decide.”

Checking Current Effort

Two ways to check:

/effort

Or:

/effort status

This displays the current effort level and its source (manual setting, environment variable, or model default).

You can also check the status bar anytime — the effort icon is always visible there.

How Effort Affects API Calls

This gets slightly technical, but understanding it helps you make better choices.

When you set an effort level, Claude Code includes an effort parameter in API requests:

• low/medium/high/max: Passed to the API via the output_config.effort parameter
• The API adjusts the model’s reasoning depth based on this parameter

In simple terms: higher effort means Claude invests more computational resources in its “thinking” phase, producing longer reasoning chains but consuming more tokens.

Persistence and Scope

Effort settings are persisted:

• low/medium/high: Saved to settings.json, effective across Claude Code restarts
• max: Current session only (unless you’re an Anthropic internal user)

Environment variable override: You can also set effort via environment variable:

export CLAUDE_CODE_EFFORT_LEVEL=medium

The environment variable takes priority over settings.json. Set it to auto to reset to the model default.

Practical Usage Tips

Daily Development: medium

In most cases, medium is the best choice. Anthropic officially recommends medium for Opus 4.6 users — it strikes the best balance between speed, quality, and rate limits.

/effort medium

Simple Tasks: Temporarily Switch to low

When you need to make a batch of simple changes (bulk renames, adding comments, formatting), switching to low can speed things up significantly.

/effort low
Rename all camelCase variable names to snake_case in this file
/effort medium

Remember to switch back after simple tasks.

Complex Analysis: Upgrade to high or max

For complex bugs or scenarios requiring deep analysis, go with high. If you’re on Opus 4.6 and the task is exceptionally complex, use max.

/effort high
This concurrency bug only appears under high load, help me find the root cause

Not Sure Which to Use?

Start with medium. If the answer feels too shallow, upgrade to high. If it feels too slow, drop to low.

The core idea of effort: allocate on demand — don’t waste, don’t skimp.

effort vs /fast

You might wonder: what’s the difference between /effort low and /fast?

• /fast: Switches to fast output mode (same model, faster output generation) — affects output generation speed
• /effort: Adjusts Claude’s reasoning depth — affects how thoroughly it thinks

They can be combined. For example, /fast + /effort low is the fastest possible combo, ideal for bulk simple tasks.

Final Thoughts

The essence of /effort is giving you control over AI’s thinking investment.

Previously, you could only passively accept Claude’s “thinking intensity” — sometimes too much, sometimes not enough. Now you can tune it based on actual task complexity: low for simple tasks, medium for daily work, high for complex tasks, max for extreme scenarios.

Use effort wisely — speed up simple tasks while ensuring quality on complex ones.