Tip: Fully Disable Thinking on Mechanical Tasks With MAX_THINKING_TOKENS=0

Claude Code Docs·medium signal

As of 2.1.166, setting `MAX_THINKING_TOKENS=0` (or `--thinking disabled`) fully turns off extended thinking on think-by-default models — previously these flags didn't reliably suppress it. For high-volume mechanical work (mass renames, formatting passes, simple edits) where reasoning tokens add latency and cost without improving output, disabling thinking trims spend. Reserve thinking budget for the genuinely hard reasoning steps in your pipeline.