Effort levels tune how much reasoning Claude puts into each turn. Low is fast and cheap, max is slow and thorough, and there are three settings between.

What it does

Each level (low, medium, high, xhigh, max) adjusts the amount of internal reasoning before Claude commits to a response. Higher levels catch more edge cases and produce better plans on complex work. Lower levels ship faster and cost less for tasks where speed beats depth.

When to use it

low for trivial edits, one-line fixes, or rote tool calls.
medium for normal day-to-day work.
high or xhigh for architecture, tricky refactors, and debugging.
max for the hardest problems where a wrong answer is costly.

Gotchas

Higher effort means slower turns. Don't use max for everything.
Effort scales tokens, which scales cost. Watch the spend on xhigh and max.
Swapping mid-task can produce mixed quality. Pick a level for a session and stick with it unless a specific step needs more.

Official docs: https://code.claude.com/docs/en/model-config.md#adjust-effort-level