Google DeepMind released Genie 4, its latest world model, in research preview on October 27, making it available to trusted testers1,2. The model is powered by Gemini 3.8 Flash and generates interactive simulations with native real-time audio, including music, ambient noises, and interactive dialogue produced directly by the model3.
Genie 4 can maintain a consistent scene for up to two minutes, double the limit of its predecessor, Google Genie 3, which entered research preview roughly a year earlier. Users can type prompts mid-simulation to alter the scene in real time, changing weather, camera angle, or adding new people and objects. The model also delivers improved physics and more human-like characters compared to Google Genie 3.
The announcement, authored by Raia Hadsell, Diego Rivas, and Hang Qi, describes Genie 4 as natively multimodal. Gemini Omni anchors simulations in real-world appearance and behavior, while other Gemini models supply the reasoning layer that maintains coherence and continuity across a session. Google DeepMind said the model "unlocks unprecedented realism and improves real-time interactive simulation".
Google DeepMind frames world simulation as central to its path toward AGI, stating that Genie 4 advances its ability to create safe, variable simulations that can support training environments for robots and agents as well as new interactive experiences.
The release arrives as the world-model sector is both accelerating and consolidating. Runway released GWM Worlds 2, a playable world model with audio generation that responds dynamically to user inputs. On the M&A side, World Labs, founded by Fei-Fei Li, is being acquired by AMD; Amazon is in talks to buy Decart; and Rivian acquired Third Dimension to bring simulation capabilities in-house for self-driving testing.
Genie 4 also lands on the same day Google described Gemini 4 Argon as its most intelligent model to date[1].
ANALYSIS Pairing a flagship reasoning model with a new world-simulation model in a single news cycle concentrates attention on Google DeepMind's breadth across both language intelligence and physical-world modeling.