GE 2.0 is an embodied world model system for robot training and evaluation, comprising the GE-Act world action model and the GE-Sim 2.0 world simulator. Given robot actions, the system generates multi-view predictions of future scenes while simultaneously forecasting joint angles and gripper states. Its built-in World Judge evaluates task outcomes against natural-language instructions and generates reward signals, enabling robot policies to be evaluated, refined through reinforcement learning, and iteratively improved within model-generated environments. This reduces reliance on physical robots and real-world experimental setups.