Google DeepMind released Genie 4, its newest interactive world model, in a limited research preview. The update adds real-time audio and the ability to edit prompts while a simulation is running.
Google says Genie 4 can hold a scene for about two minutes — roughly double Genie 3 — though drift and audio accuracy remain known limitations.
World models sit next to chat LLMs as a second AI frontier: systems that simulate environments for games, robotics research and agent training rather than only answering text prompts.
Questions? Chat with our team — threaded comments are coming soon.