Tip: Practitioners Report Medium Effort Beating High on Opus 5 for Everyday Coding Work
The same 22 August Hacker News thread produced a repeated, independently-stated operating tip: several heavy users report far better results on medium effort than on high or above, with one describing switching to Low reasoning instead of falling back to Sonnet and getting better quality at comparable spend, because the majority of the Opus-versus-Sonnet cost overhead came from Opus being chattier rather than smarter. Another user reports a 'read and update the config file' prompt turning into 43 minutes of pulling containers, running sandboxes and building test suites for a one-file change. This is anecdotal and unverified, but it is cheap to test on your own workload and the failure mode it names is measurable.
Source
↳ Follow the thread