vLLM Semantic Router

vLLM Semantic Router has released d3-ultra, a 27B open-weight multimodal decision model. It reads text, images and video once and returns probabilities for every answer. The team reports a score of 87.2 on its SR-Text test.