Anthropic's Claude Opus 4.7 has a writing problem. Multiple Hacker News users report the newest version has lost what they call its "soul" for creative tasks. Where Opus 4.6 produced prose with some literary warmth, 4.7 gives responses that feel checked off rather than written. Terse. Flat.
The trade-off isn't accidental. Users say Opus 4.7 is noticeably sharper at logical reasoning and better at catching bugs mid-implementation. For developers, a real upgrade. For writers who relied on Claude, a clear regression.
The fix, according to multiple commenters: stick with Opus 4.6 for anything creative.
This pattern keeps showing up across AI companies. Optimize for one capability, lose ground on another. One HN user suggested the writing issues reflect Anthropic tuning Opus toward its intended purpose, and that people should adjust their approach. Fair point, but cold comfort if you picked Claude specifically because the old version wrote well.
Anthropic hasn't publicly addressed the feedback, so it's unclear whether future updates will try to balance these strengths or continue down the technical path.