Claude Code usage running out faster than expected? First, type /effort status in the chat to see which effort level this session is really using. The level you pick in the menu may not be the one that counts. The official docs say the environment variable CLAUDE_CODE_EFFORT_LEVEL comes first. If it is set, it overrides whatever you choose in /effort or the model picker.
Which setting wins
The effort level controls how deeply the model thinks at each step. A higher level means more thinking and more tokens used. The official docs say Claude Code looks for the level in the order below and uses the first one it finds:
| Order | Source | Notes |
|---|---|---|
| 1 | Environment variable CLAUDE_CODE_EFFORT_LEVEL |
If set, it always applies and cannot be changed inside the session |
| 2 | Chosen in the session | The /effort command or the slider in /model |
| 3 | Startup flag --effort |
Applies to this one launch only |
| 4 | Settings file | A level saved for each model, or effortLevel |
| 5 | Model default | Fable 5.2, Opus 5.5, Sonnet 5.5, and Haiku 5.5 default to medium |
There are three common traps:
- The environment variable can hide in a settings file. It can be set in your shell config, or in the
envsection of a settings file. Someone might set it tomaxfor a hard problem long ago, forget about it, and then every session runs at the highest level. maxusually lasts only for the current session. The official docs saymaxstays on across sessions only when you set it with this environment variable. So ifmaxis on every time you open Claude Code, this variable is the likely cause.- Subagents and Skills have their own effort levels. Their instruction files can include
effort. They run at that level, no matter what you picked for your main session.
One user report says that the Desktop app was set to medium, but a subagent used up the whole five-hour usage limit in ten minutes. The user later found an override in a local settings file. This is one person's experience and is not verified yet, but it matches the order above.
Ask your AI agent to check
Send this to your AI agent:
Help me check whether my Claude Code effort level is being overridden. Read only, do not change anything:
1. Check whether the environment variable CLAUDE_CODE_EFFORT_LEVEL is set, and in which file. Look at the shell config, ~/.claude/settings.json, and the env section of the project's .claude/settings.json and settings.local.json.
2. Check what effortLevel and modelSettings say in these settings files.
3. Check which files in ~/.claude/agents, the project's .claude/agents, and skills have effort at the top.
4. Make a table with: location, current value, and what it affects. Tell me which line to delete to go back to the default level. Wait for my confirmation before changing anything.
After the change, start a new session and run /effort status again to confirm.
Notes
- To clear the level saved for a model, use
/effort auto. - Use a lower level for simple tasks. It is faster and cheaper. Raise it only for hard problems. Writing
ultrathinkin a prompt makes that one turn think deeper without changing the session level. - If your work account is set up by an admin, the admin may have set a maximum level. In that case, picking a higher level will still run at the maximum.
Official docs: Model configuration in the Claude Code docs.


Comments (0)
No comments yet. Start the conversation.