Hacker News
'Why Does Opus 5 Feel Worse to Work With?' — 100 HN Comments on Benchmark Gains That Degrade Daily Use
A post published August 14, 2026 argues Opus 5 is objectively more capable — the author concedes it "rivals Fable in benchmarks" — yet is worse to actually work with than Opus 4.7, 4.8, or Fable, because it stops asking clarifying questions, makes unverified assumptions, and reinterprets or updates the user's plan without asking. The author offers no quantitative measurements, attributing the shift to two pressures: optimizing for self-improving systems and optimizing for benchmarks, both of which reward confident assumption over clarification. It drew 102 points and 100 comments within hours, making it one of the more contested practitioner threads of the day.
Source
↳ Follow the thread