Google DeepMind Launches Genie 4 Real-Time World Model With Native Audio

Google DeepMind announced Genie 4 on October 27, 2026, describing it as its most realistic and interactive world model. The model generates dynamic environments in real time without pre-rendered assets or hand-programmed interactions, with users providing natural language prompts to shape the experience. Genie 4 is the first version with native audio, covering ambient sound, music, and live character dialogue. Gemini Omni anchors simulations in real-world appearance, while Gemini 3.8 Flash acts as an in-loop director that breaks goals into steps and analyzes incoming frames to maintain scene coherence. DeepMind highlighted human simulation as a particular advance: characters move naturally, respond organically to their environments, and can be steered in real time. Gemini 4 Argon used the Genie 4 loop to simulate Apptronik's Apollo humanoid robot playing checkers. All output is watermarked with SynthID. Access is rolling out via Project Genie to Google AI Ultra subscribers in the U.S. (18+, not available to Google AI Ultra for Business users), with plans to expand to more territories and to Gemini API early access partners.