DeepMind introduces Genie 4 world model with enhanced physics and audio

Google DeepMind has announced Genie 4, an updated world-model system that builds on the capabilities of its predecessor, Genie 3. The new model generates virtual environments in real time, allowing users to modify parameters such as weather, objects, or camera angles while the system maintains contextual continuity.

Genie 4 relies on the Gemini 3.8 Flash model to break down user goals and evaluate each output frame, helping keep simulations on track. While the system can sustain consistent worlds for only a few minutes before resetting, it introduces several technical advances, including more accurate aerodynamics, collision handling, and less uncanny human movement.

For the first time, DeepMind’s world model incorporates native audio, enabling ambient sounds and user-specified audio cues. The company also claims the model can produce interactive human speech that adapts to changes in the simulation.

Access to Genie 4 will initially be offered to developers through the Gemini API and AI Studio. Consumer availability is planned for Google’s AI Ultra subscription tier, which begins at $100 per month, with potential expansion to other plans later. A waitlist is open for early testers, though a detailed rollout schedule and pricing have not been disclosed.