Google DeepMind has introduced Genie 3, its most advanced “world model” to date. The system can create navigable, fully interactive 3D environments from a single text prompt, rendering them in real time at 720p and 24 frames per second. Genie 3 keeps track of prior frames for roughly one minute, allowing objects, lighting and physics to remain coherent for several minutes of exploration. Users can also trigger “promptable world events”—such as changing the weather or adding new characters—without restarting the simulation. The model marks a substantial leap over December’s Genie 2, which was limited to 360p output and about 10-to-20 seconds of stable play. In an internal test, DeepMind’s generalist SIMA agent successfully executed instructions inside a Genie-generated warehouse, underscoring the system’s potential for training embodied AI. DeepMind says the technology could streamline game development, provide infinite training grounds for robots and serve as a step toward artificial general intelligence. For now, Genie 3 is available only in a restricted research preview to select academics and creators while the company studies safety, scalability and longer-duration performance.