I ran Claude Code at all 5 effort levels on one task #ClaudeCode

I ran the same bug fix in Claude Code at low, medium, high, xhigh and max on Opus 5.5. Low took 27 seconds. Max took about 25 minutes and used about 17 times the tokens of medium. Every level made the same fix. The three moves: 1. Keep medium as your default on Opus 5.5. 2. Go to max only to plan a very complex system, and for its first run. 3. Once you like the output, switch back to medium. Which level do you keep as your default, and have you ever changed it? Effort levels in the docs: https://code.claude.com/docs/en/model-config #Shorts #ClaudeCode #AIWorkflow