Meta’s V-JEPA 2 is a new video-trained world model designed to help AI agents and robots predict physical outcomes before acting—much like humans use intuition to navigate, avoid obstacles, or anticipate motion. Building on its predecessor, V-JEPA 2 enhances understanding, prediction, and planning by learning how objects and people interact through vast video data. The result: robots powered by V-JEPA 2 can now perform tasks like reaching, picking up, and placing objects—even in unfamiliar environments—pushing AI one step closer to advanced machine intelligence.

Meta