Google DeepMind has released Genie 4, an updated world model that builds on the success of Genie 3 by incorporating improvements driven by its updated Gemini models. When prompting a world into existence, Genie 4 leverages Gemini 3.8 Flash as a guiding agent to decompose goals and evaluate each output frame, helping maintain simulation coherence across varying environments while preserving original context. Users can also change parameters midstream—such as weather conditions, add elements, or move the camera—with Gemini providing real-time steering updates after each modification, analogous to how reasoning models generate additional internal prompts to refine output tokens. Despite these advances, DeepMind acknowledges that Genie 4 worlds still suffer from context degradation: simulations remain consistent only for approximately two minutes before losing their initial state. World models, in general, generate virtual environments frame-by-frame in real time without pre-rendered scenes or programmed interactive elements, enabling infinite training environments for physical robotics and rapid prototyping for creative fields. The release reflects DeepMind's broader strategic focus on advancing toward artificial general intelligence through simulation technologies that can dynamically model physical and social worlds.