A single unified model that performs multiple tasks—proposing actions, generating predicted images, and reasoning—rather than separate specialized models.