Users Flag Verbose Claude Responses and Tokenizer Shift
Social media users flag recent shifts in Claude response style and possible model updates.
TLDR
Kim Isenberg posted that Claude now answers in convoluted, overly long, and wordy terms that make the experience no fun. Prompt engineer Nick Dobos replied that a tokenizer change arrived in 4.7. He also stated that Anthropic began distilling Mythos and Fable models into Sonnet and Opus. The posts cite visible differences in output length and phrasing. No first-party confirmation from Anthropic appears in the packet. The remarks remain user observations shared on X.
Combined views
37.4K
2 Sources, first seen ago
Sources
ST 
Samy TOURI@samy_touri@kimmonismus It recently gave me the filling of "am I becoming stupid or something? I read without understanding a single word and losing the context myself" happy to hear I am not the only one because it is extremely irritating! (I stopped my Claude subscription, planning to…
BN Borislav Nikolov@Borisla30071961@Priebstar @kimmonismus That seems like a likely contributor, whatever the underlying cause may be. Also, selecting different tokens for watermarking could itself affect the stop condition and lead to longer outputs, even if increasing output length is not the goal.
CS Chahat Sharma@Chahatusharma@kimmonismus Claude's default verbosity is genuinely painful. You ask a yes/no question and get four paragraphs.