According to ViggleAI via X, the company released JST-2, a full-stack physics-native world foundation model that builds in 3D rather than pixels and generates expandable 3D scenes and motion-ready characters in seconds.