Anthropic engineer Jackson Kernion explained why new Claude models, according to The Decoder's description, have started writing in an unfamiliar way. Optimization for mathematics, programming, and technical explanations for other AIs has led to a style that people may perceive as overly dense blocks of information.
The article states that Opus 5.5 attempts to correct this feature, while Opus 4.6 remains the strongest model for purely writing tasks. The source provides no details on the comparison methodology and is based on a publication synopsis rather than the full text.
For users, this highlights a practical trade-off: increased technical capabilities of a model do not necessarily make its responses clearer or more natural. The question of how widespread and measurable this effect remains open.
editorial commentary
Why it matters
A likely consequence is greater attention to clarity and naturalness of text as a separate quality criterion for AI. The next observable signal will be independent comparisons of Opus 5.5 and Opus 4.6 on writing tasks. Significant uncertainty remains because the findings are presented via a single The Decoder synopsis without the full text or additional confirmation.