A visualization of the different effort levels for Claude models.

I find myself using medium for 80% of the tasks, high for ~15%, and occasionally will reach for xhigh or max when working on something super complex, but Opus 5.5 on medium I find perfect for most coding work.