Hacker News
Anthropic's Claude Code Team Publicly Answered the 'Reduced Effort Levels' Accusation on a 195-Point HN Thread
A claim that Anthropic was A/B testing lowered effort levels in Claude Code reached 195 points and 173 comments, with users reporting that a high effort setting self-reported as '10' and that Opus 5 spent 43 minutes pulling containers and building test suites on a task version 4.6 finished in under two minutes. Thariq from the Claude Code team replied in-thread that one running experiment maps the numerical effort value differently, that the displayed number is not meaningful, that 'the effort you selected is the effort you're getting,' and that in-depth evals confirmed no performance impact, offering credits for demonstrated regressions filed via /feedback.
Source
↳ Follow the thread