Models & Architectures
Decision Model in plain English.
Also known as: classifier model,single-pass decision,routing model
The one-sentence version
A small model that returns an answer or label in one forward pass instead of generating text, built for routing, verification, and fast yes-or-no choices.
A decision model does not write; it decides. Given a question, an image, or a snippet and a fixed set of possible answers, it returns a choice (and often a calibrated probability) in a single forward pass, with no token-by-token generation. That makes it fast and cheap enough to run on every request: Liquid AI's d1-3B, released as open weights on October 9, 2026, answers in about 8 ms on a consumer GPU by the company's figures, and its d1-omni-600M variant takes audio or images. Decision models sit around language models rather than replacing them: routing requests to the right model, verifying an agent's output before it acts, moderating content, classifying intents, and gating tool calls. OpenAI's Decisions API, announced at DevDay 2026, offers the same shape as a hosted service. The limits are the fixed answer set and the dependence on training data for the domain; a decision model cannot explain itself the way a generative model can, so teams pair it with logging and occasional generative review.