Where would a 144M yes/no model replace a 30B chat call in your stack?

The newest local-LLM wave isn't another chat model — it's models that never write prose at all....