The big picture

Capable multimodal reasoning has been largely cloud-bound. These models shift that balance, putting vision and decision-making on local devices.