Liquid AI Releases d1: A Decision Model That Returns Calibrated Probabilities With Zero Output Tokens

A decision model is a type of AI built for making bounded choices rather than generating text. It takes context and typed questions, then returns calibrated probabilities across predefined outcomes in a single call with zero generated output tokens. This approach differs from general language models because it produces no text output, delivers predictable latency without a decoding loop, eliminates schema errors, and provides usable uncertainty through calibrated probabilities rather than self-reported confidence scores. Decision models are best suited for tasks like classification, routing, moderation, scoring, and reranking, while language models remain better for summarization, drafting, and multi-step reasoning.