Liquid AI kills LLM overkill

🔥 What happened Liquid AI shipped d1, a decision model that returns typed answers with calibrated probabilities instead of generating text. Every call burns exactly 0 output tokens. It's live today as a hosted API under the name d1:free. 💡 Why it matters Most teams still route classification, moderation, and LLM-as-judge work through general LLMs — which bill output tokens even for a one-word label. d1 offers three primitives (Noul, Choice, Score), batches multiple questions into one round trip, and makes thresholds like 0.8/0.2 actually usable. ⚡ Our take If you're still burning GPT calls on ticket routing in 2026, you're paying twice: in dollars and latency. d1 isn't an LLM killer — it's the death of the "one model for everything" reflex.