D3-mini

D3-mini is a compact multimodal decision model in vLLM Semantic Router’s Decision 3.0 family. It combines a 4B-class backbone with a vision encoder to answer choice, yes/no, and scoring questions about text, images, and video. It returns probabilities over defined answers rather than generated text.⁠

Use cases

Use it to classify incoming requests, route work based on text and attachments, check receipts or images against defined conditions, or score urgency and relevance. Its smaller model size makes it a candidate for deployments where resource requirements matter alongside decision quality.

Inputs and outputs

D3-mini accepts text or JSON with optional images and videos, including several media items per request. You can ask multiple questions about the same input and receive structured decisions with probabilities for each allowed answer.⁠

Deploy on Baseten

Use the Deploy button to start from the Baseten deployment template. Use the deployment URL and authentication instructions from your Baseten dashboard.

Model card and local usage example

About vLLM Semantic Router