Starting from 0.6B! Jev's rival model arrives, open-source multimodal, tops the global evaluation leaderboard

What happened: vLLM Semantic Router released and open sourced the Decision 3.0 multimodal decision model series, which returns judgments and probabilities instead of long text. Key numbers: Six models from 0.6B to 27B parameters. d3 scored 64.7 on the Jev Decision Index 0.3.1 text leaderboard, first place, above Jev's 60.1. d3 scored 71.6 on the vision leaderboard, also first. Why it matters: The open weights support local deployment and direct image and video handling, unlike Jev's cloud API. d3, d3-flash and d3-mini all beat Microsoft Decision-1's 53.2 score.

vLLM Semantic Router released and open sourced the Decision 3.0 multimodal decision model series, which returns judgments and probabilities instead of long text.

The open weights support local deployment and direct image and video handling, unlike Jev's cloud API. d3, d3-flash and d3-mini all beat Microsoft Decision-1's 53.2 score.