RoboJev-27B for Open-Weight Robot Success Detection
October 23, 2026
RoboJev-27B is a 27B parameter vision-language model specifically trained to predict robot task success from video frames. It uses a calibrated two-token readout to provide success/failure probabilities in a single forward pass.
HOW THIS AFFECTS YOU
●
builderYou can integrate this model into your robotics pipeline to automatically evaluate task completion.
●
researcherThis provides an open-weight alternative for evaluating robotic manipulation performance.