Tag
RoboJev-27B is an open-weight vision-language judge that predicts robot manipulation success from video frames, achieving 95.2% balanced accuracy on the SO-101 robot and strong generalization across unseen labs and robots, with inference code and full evaluation data released.