jevIMAGE
AutoTrust 开源多模态决策模型 JEV-27B-VL 下载破 154 万
该模型输入屏幕、图像和文本,输出带校准概率的动作选择,用于 computer use、推荐和 agent 评估,在 Jev 决策榜上 13695 例准确率 77.8%。
AutoTrust
@AutoTrustAI19 天前收录原推发布于 2026.10.11

AutoTrust @AutoTrustAI原推发布于 2026.10.11
Thank you @HuggingModels for featuring AutoTrust AI’s JEV-27B-VL! From 179K downloads in this post to 1.54M today. Our open multimodal decision model also leads the Jev Decision Index vision board: 77.8% accuracy across 13,695 examples. Screens, images, and text in. Calibrated probabilities over actions out — for computer use, recommendation, and agent evaluation. Download the model: https://huggingface.co/autotrust/JEV-27B-VL Explore our research workspace. Download ScienceGuru: https://scienceguru.ai/
这条是引用着另一条写的:x.com/huggingmodels/status/2106131865589690376
x.com/autotrustai/status/2109156672698786002采集于 2026.10.12 10:05