Decision 3.0: open multimodal decision models (0.6B-27B)
vLLM Semantic Router team's new family of six Apache-2.0 decision models that answer typed choice/yes-no/score questions over text, images, and video with probabilities instead of generated prose.
Why it matters
First open decision-model family that is multimodal out of the box (reads images and video zero-shot) and tops the Jev Decision Index boards; sub-second decisions on a single GPU make it a cheap, auditable gating layer for agentic pipelines.
What you could build with it
An indie developer building a customer-support triage bot could use Decision 3.0 as the routing layer: every incoming ticket with screenshots gets classified (billing vs bug vs how-to) and scored for urgency in one call, so the right specialist agent handles it without an LLM writing and parsing prose first.
Does it hold up?
Too early to judge: weights shipped Oct 10 with self-reported benchmarks only; no independent reproductions or production deployments are public yet. The Apache-2.0 license and one-line vllm-sr serve command make it easy to try.
Built with Decision 3.0
- Decision 3.0 Is on Hugging Face: Six Open Multimodal Models (OrcaRouter)blog · Independent writeup documenting the six checkpoints, their Oct 10 HF drop, and the caveat that all performance claims are self-reported and unaudited.
- What is Decision 3.0? (MarkTechPost)article · Comparison table of d3 vs Perplexity Decider, Decision 2.0, and LiquidAI d1 across parameters, context, latency, and decision-index scores.
- Team announcement with 40 comments (LinkedIn)linkedin · Maintainer's launch thread: 27B ranks #1 on both Jev Decision Index boards, 9B d3-flash beats Decision 2.0's 27B, video QA out of the box.
- vllm-project/semantic-routergithub · 6,173 ★ · vLLM Semantic Router serves Decision 3.0 natively in Engine mode behind the System One API; the models are its open decision layer.
Learn more
First spotted on hf: source.
More AI models
OpenAI publishes 722 AI-generated math manuscripts from an unreleased modelOpenAI released 722 mathematical manuscripts in 372 result families to GitHub under Apache-2.0, produced by an…models · JEV 0.74StepFun Step-5-Preview — open BF16 weights releasedStepFun published the full BF16 weights for its flagship Step-5-Preview agentic model (600B-parameter sparse MoE, 27B…models · JEV 0.73ElevenLabs Eleven v4 + v4 Turbo: new emotive TTS architecture with a 100ms real-time variantElevenLabs released Eleven v4 and Eleven v4 Turbo on September 28: a new text-to-speech architecture with inline…models · JEV 0.72Mistral Large 4 (Le Chonk): 1.05T-param open-weight MoEMistral's new flagship: a 1.05T-parameter mixture-of-experts multimodal model with 49B active params and 1M-token…models · JEV 0.72VoiceStudio: open-source, fully-local ElevenLabs alternative with voice cloning, dubbing and dictation in 646 languagesVoiceStudio is an open-source, fully-local voice AI studio: voice cloning from a 3-second sample, voice design, video…models · JEV 0.71JetBrains Mellum 2.1: 12B MoE coding model trained for agentsJetBrains released Mellum 2.1, a 12B MoE model (2.5B active parameters, 128K context) under Apache 2.0, trained mainly…models · JEV 0.69
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.