Claude Opus 4.8 and 5.0 criticized for incoherent and toxic language style

User Complaints Regarding Claude Opus Language

Recent versions of Claude Opus, specifically 4.8 and 5.0, are drawing criticism from users for their output language. Reports indicate that the models' default writing style has become verbose, filled with jargon, and overly stylized, making the generated text difficult to comprehend.

A Reddit thread discussing these issues garnered over 450 upvotes and 175 comments, with users largely agreeing on the presence of serious language calibration problems in Opus 4.8. This sentiment suggests a widespread dissatisfaction among the user base.

Impact on Readability and User Experience

Users describe the model's responses as padded with invented corporate/hype jargon, forced metaphors, and 'catchy' phrasing that obscures the actual meaning. This forces users to re-read sentences multiple times to extract information, and some have resorted to piping Opus output through other models for simplification.

The perceived tone is often described as trying to sound clever rather than being clear, leading to frustration, especially when the generated information is incorrect. This issue affects various applications, including the chat/web UI, Claude Code, and Fable 5.

Recurring Stylistic Patterns

Specific recurring patterns in the model's output include the frequent use of unusual terms like 'Load bearing,' 'prose' (instead of text), 'Hand-waving,' 'Reflexive hedging,' and 'Honest framing.' The model also tends to lead sentences with what something is not, rather than what it is.

Additionally, users note the invention of jargon and aphorisms presented as standard, such as 'instrumentation is the unlock,' and the use of forced metaphors that require decoding rather than aiding understanding.