D3 Ultra

d3-ultra is the largest model in Decision 3.0, the open decision model family from the vLLM Semantic Router project. You send a state (text or any JSON value) and named questions, and it returns typed answers: the probability that a statement is true, a choice among options you define, or a position on your own rubric, each with a full probability distribution. It reads the state once and answers every question from that one read, and it generates no text, so answers cannot fall outside the options given. It speaks the same request and response format as TypeSafe's System One API, so code written for Jev can switch models by changing the model name. On its authors' own text decision benchmark it scores ahead of the smaller d3 models and Jev 1.14. Use it where accuracy matters more than raw speed: routing and triage, guardrail checks, grading and extraction-style questions over long documents. The weights also support images, video and box answers; the endpoints on Opper today take text and JSON states. Weights are published under Apache-2.0 at https://huggingface.co/vllm-sr/d3-ultra.