Google DeepMind unveils Genie 4 world model with native audio and real-time steering

Google DeepMind has released Genie 4, its newest world model, as a research preview for trusted testers. The company announced the model on Oct. 27, 2026, calling it a step toward "unprecedented realism" in real-time interactive simulation.

What is new

• Native audio: Genie 4 is DeepMind's "first real-time world model to generate audio natively", from ambient sound to interactive dialogue that adapts to user inputs, according to the DeepMind blog.
• Real-time steering: users can change the weather, move the camera or add a new person or object with natural-language prompts while the simulation runs.
• Gemini in the loop: Gemini 3.8 Flash observes the scene, breaks goals into steps and reviews incoming frames to verify each step, DeepMind said.
• Longer scenes: experiences can run about two minutes, up from one minute for Genie 3, The Verge reported. TechCrunch called it double its predecessor's limit.

Robots, film and games

DeepMind also showed Gemini 4 Argon breaking a task into steps that Genie 4 simulated in real time, with Apptronik's Apollo humanoid playing checkers. Trusted testers are using the model for film pre-visualization, game prototyping and scenario simulation, the company said. "What we'll increasingly see in the future are these hybrid world models," Google research director Tejas Kulkarni said in a briefing quoted by TechCrunch.

Limits and safeguards

DeepMind said visual details and object permanence can still drift over longer sessions, and that audio accuracy and legible in-world text remain open research problems. All video and audio generated by Genie 4 will include SynthID watermarking, the company said. TechCrunch noted the launch comes as the world-model field consolidates, noting that Fei-Fei Li's World Labs is being acquired by AMD.