Yes but because it's trained on text and in the backend, images are converted to tokens, it is absolutely dogshit at navigation and basic puzzles. It can't figure out what Squirrels can about how to achieve goals in a maze.