You can show this model an image or a video, and it will answer any questions about it. It goes way beyond what frontier models can do. For example, Glimpse can tell you how objects in an image relate to each other, or how an action changes over time in a video. Robotic teams would kill for these capabilities!