What is Genie 4?
Genie 4 is the fourth generation of Google DeepMind's world model, announced on October 27, 2026. Like its predecessor, it generates an interactive world in real time that you can walk through and change while it runs. What is new is that the world now sounds and is populated: Genie 4 produces music, ambient noise and spoken dialogue together with the picture, and the people in the scene move naturally, react to their surroundings, talk and can be steered as the simulation unfolds.
DeepMind describes it as natively multimodal: you set an event in motion, and the model simulates how the world responds, how it sounds, and how people move and speak.
What's new compared with Genie 3
Genie 3, released in August 2025, generated explorable 3D-looking worlds at 720p and 24 FPS and kept them consistent for minutes at a time, with promptable world events. Genie 4 keeps that idea and adds four things: native audio, steerable characters, sharper photorealism with more fluid movement, and physics scenarios such as aerodynamics and collisions.
The most important architectural change is that Gemini is built into the loop. Gemini Omni anchors how the world looks and behaves, while other Gemini models reason about coherence and continuity. When you change the weather, move the camera or add people and objects with a text prompt, Gemini adapts the edit to the surrounding scene and keeps the world progressing between your inputs, instead of waiting for the next command.
Agents inside the world
Genie 4 is also designed as a place where AI agents can act. In DeepMind's examples, Gemini 3.8 Flash observes a Genie scene, breaks a goal into steps and checks whether each step succeeded. In a separate experiment, DeepMind paired Gemini with Genie 4 to simulate Apptronik's Apollo humanoid robot playing checkers — a glimpse of world models as training grounds for robots.
Specs and limitations
Google has not published Genie 4's resolution, frame rate, latency, model size or maximum session length. Its model page speaks only of “a few minutes of interaction”. You may see “two-minute scenes” in launch coverage, but that figure does not come from Google's own materials.
DeepMind is candid about what still breaks: visual details and object permanence can drift over time, especially in long sessions, after several prompt edits, or while navigating. Complex physical interactions, real-time audio accuracy, legible text, spatial accuracy and long-term consistency remain open research problems.
How to get access
Genie 4 is not publicly available yet. It is rolling out as a research preview to trusted testers, then to Google AI Ultra subscribers through Project Genie and to early-access partners of the Gemini API and Google AI Studio. A waitlist is available through Project Genie. Everything it generates carries SynthID watermarking.
Strengths and limitations
Strengths
- Native audio, including dialogue, generated with the picture
- Characters that move, react, speak and can be steered
- Gemini keeps edits coherent and the world progressing between inputs
- Built for agents and robotics simulation, not just play
Limitations
- Closed model; access limited to testers and Ultra subscribers
- No published resolution, frame rate or session length
- Object permanence can drift in long or heavily edited sessions
- Cannot be run, studied or fine-tuned on your own hardware
Who is Genie 4 for?
Anyone who wants the most advanced real-time world simulation available today and has, or can get, access through Google AI Ultra or the Gemini API — and researchers interested in agents and robots acting inside simulated worlds.
Frequently asked questions
When was Genie 4 released?
Google DeepMind announced Genie 4 on October 27, 2026, as a research preview for trusted testers, with a rollout to Google AI Ultra subscribers and Gemini API early-access partners.
How is Genie 4 different from Genie 3?
Genie 4 adds native audio and dialogue, characters that move and speak and can be steered in real time, sharper photorealism, and Gemini models built into the loop to keep text edits coherent and the world progressing between inputs.
What resolution and frame rate does Genie 4 run at?
Google has not disclosed it. For reference, Genie 3 ran at 720p and 24 FPS.
Is Genie 4 open source?
No. Genie 4 is a proprietary Google DeepMind model available only through Google's own products and APIs.
How can I try Genie 4?
Join the Project Genie waitlist. Access is rolling out to Google AI Ultra subscribers and to early-access partners of the Gemini API and Google AI Studio.