Benchmarks & research
Decision 3.0 is an open-weight family of multimodal models ranging from 0.6B to 27B parameters that can classify text, images, and video.
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
There are two different Qwen bases here: Qwen 3.8 for the flagship 27B model and Qwen 3.5 for the 9B flash variant. The flagship scores 66.26 on the full Jev Decision Index 0.3.2, compared with Jev 1.13.0’s 60.11; its public-suite score is 65.45. JevMade has not independently measured these results.
Pictures or video can accompany the situation and list of possible answers you give the model. It scores those answers without composing a reply. This is vLLM Semantic Router’s family, not Jev; choose among its sizes according to your computing needs.