decision-index-vision-leaderboard

# Decision Index Vision The vision board of the [Jev Decision Index](https://huggingface.co/spaces/multimodalart/jev-decision-index): decisions over images and screens. Text board: [multimodalart/decision-index-leaderboard](https://huggingface.co/datasets/multimodalart/decision-index-leaderboard). Tasks (each is its own leaderboard): - `decision_index_vision`: Full = mean of the public and private vision scores. - `public`: the 11 public vision benchmarks. - `private`: the private vision sets (not released; only aggregate scores are published). Images are capped at 1.6 MP for every model. Methodology: https://huggingface.co/spaces/multimodalart/jev-decision-index. This repo mirrors the board: `leaderboard` has one row per model (62 entries), `benchmarks` the per-benchmark public scores. It is regenerated whenever the board is updated. ## Top 10 | Rank | Model | Full | |---|---|---| | 1 | WeMM-JEV-27B | 73.81 | | 2 | RBDash-Decision v0.2 | 73.30 | | 3 | vLLM-SR d3-turbo v1.1 (27B) | 72.46 | | 4 | Amos 1 Large (27B) | 72.40 | | 5 | AutoTrust-JEV-27B v1.1 | 71.89 | | 6 | Perplexity Decider v1.2 (27B) | 71.72 | | 7 | StartLux-Decision v1.1 27B | 71.61 | | 8 | Fastino GLiDE 1.2 | 71.46 | | 9 | VisionJEV 1.0.0 (27B) | 71.42 | | 10 | vLLM-SR d3 (27B) | 71.26 | Generated 2026-10-29T14:04:16Z.