vLLM Semantic Router releases Decision 3.0: six Apache-2.0 multimodal decision models
Decision 3.0 (d3) is six Apache-2.0 decision models, from d3-edge at 0.59B to d3 at 26.09B parameters. Give one text or JSON, optionally with images or video, plus questions (pick an option, yes/no, or a score), and it returns a probability for every answer in one call without generating text. The team says d3 ranks first on the Jev Decision Index text board with 66.3; that is the team's own claim and has not been independently replicated.
How you’d use it: Route, screen or score requests in one pass instead of asking a chat model.