Should You Run Decision 3.0 Locally?

On October 10–11, 2026, vLLM Semantic Router shipped Decision 3.0: six Apache-2.0 multimodal decision models on Hugging Face (d3-edge ~0.59B through d3 ~26.09B). They answer choice, yes/no, and score questions with probabilities—no generated text. The model card reports a 55 ms text median on one AMD Instinct MI325X and a vendor Jev Decision Index of 64.7. Independent coverage (OrcaRouter, Upstream Beat) confirms the weights are real and flags that the leaderboard numbers are not yet independently reproduced. Comment OPEN for a practical checklist. Watch the Short:    • 55 ms Local Decisions: Decision 3.0's Catc...   Chapters: 0:00 INTRO 0:11 WHAT 1:03 NUMBERS 2:11 CATCH 2:57 WHO 3:14 DO THIS Sources: https://huggingface.co/vllm-sr/d3 https://huggingface.co/collections/vl... https://x.com/XunzhuoLiu/status/21092... https://www.orcarouter.ai/blog/decisi... https://upstreambeat.com/news/vllm-se... Subscribe for new AI releases explained without the hype. #Decision30 #vLLM #OpenSource #HuggingFace #AIAgents #Multimodal