Aug 20, 2026 16:16 UTC
hacker news
https://news.ycombinator.com/item?id=49375996
A discussion is gaining traction on Hacker News about using a separate LLM to clean up perceived 'token vomit' from Claude 5.
READY TO POST:
seeing some chatter on hn about claude 5's verbosity and people building tools to tidy up its output with another llm.
is claude 5 a bit too verbose for anyone else? a new project aims to 'clean up' its token vomit using a second llm.