Opus 5 feels worse to work with, and benchmarks may be why

Why does Opus 5 feel worse to work with?

Despite being more capable on benchmarks, Opus 5 feels like a downgrade to many developers compared to Opus 4.7, 4.8, and Fable. The author argues that these models are better because they ask clarifying questions and avoid making unverified assumptions. This preference for clarification is at odds with the training incentives of frontier labs, which reward models that make bold assumptions to score well on self-contained benchmark tasks. Real-world coding, however, is full of ambiguity, making such behavior risky and requiring constant babysitting.