RCreddit.com
12
·20 hr ago·Dev community · RSS
My RULE of Thumb of choosing a models
Model release
Heat trend
New
The percentage is based on available heat signal, not comment count or independent people.
This is mostly for setting up for expectation, since personally without LLM i could take 3 days (15 hours of active programming) to debug or implement a feature, but with Qwen 27B (even before Qwen 3.8) it take 4 hours.
And yes 0.5 tok/s is human, not accounting of deletion and pausing, that's also the reason i am fine leaving overnight code base wide analysis or fin tech and deep research.