Many MindedDeepMind's Genie 4 simulates steerable worlds with sound, for testers only
Google DeepMind announced Genie 4 on October 27, 2026. It is a world model: software that generates an interactive simulated environment you can move through. The announcement was written by Raia Hadsell, Diego Rivas and Hang Qi. It is a research preview for trusted testers, with feedback gathered before a wider release.
According to DeepMind, the model is natively multimodal and produces audio, including music, ambient noise and interactive dialogue, at the same time as the world itself. It builds on the Gemini family. Gemini Omni grounds the simulation in how things look and behave in the real world, and other Gemini models supply reasoning to keep the scene coherent. Users steer with natural-language prompts, such as changing the weather, moving the camera or adding a person or object. They navigate with standard keyboard controls. DeepMind says physics like aerodynamics and collisions are simulated, and that its biggest improvement is people who move naturally and respond to steering.
The abundance angle is training environments. Robots and software agents learn from huge amounts of practice, and real-world practice is slow and costly. If a simulator can produce varied, believable scenes on demand, that practice gets cheaper. Interactive learning content is a second use, which touches knowledge. This is our analysis. DeepMind itself says only that Genie 4 can help create training environments and new interactive experiences.
What to watch: the planned release starts with Google AI Ultra subscribers in Project Genie, plus early-access partners of the Gemini API and Google AI Studio. A waitlist is open. No general release date is given. The source is DeepMind's own announcement, so its realism claims are unverified by third parties. The available text includes no benchmarks, resolution, frame rate, latency or session length, and it is truncated, so some capability sections are not covered. Whether the cost and access reach beyond a premium tier will decide how much this matters.
Source: Google DeepMind