# Google DeepMind unveils Genie 4, a world model with real-time audio

> The research preview generates interactive scenes with sound and can be steered by prompt, but only trusted testers can use it for now.

URL: https://claudegpt.si/news/genie-4-world-model-preview  
Published: 2026-10-27T18:52:35.771Z  
Author: ClaudeGPT Si Newsroom  
Section: AI Models  
Topics: Google DeepMind, Genie, World models, Gemini  
Publisher: ClaudeGPT Si

## Key takeaways

- Google DeepMind announced Genie 4 on October 27 as a research preview for trusted testers.
- It generates audio natively with the visuals and can be steered by natural-language prompts.
- DeepMind has not stated resolution, session length, pricing or a general release date.

Google DeepMind announced Genie 4 on October 27, a world model that generates interactive, real-time simulations with sound. It is a research preview for trusted testers only, according to [DeepMind's post](https://deepmind.google/blog/genie-4-a-new-generation-of-world-simulation/).

A world model is an AI system that generates a playable, changing scene rather than a fixed video. You steer it while it runs. That makes Genie 4 useful for testing agents and game ideas, and it is why DeepMind keeps pointing at simulation rather than entertainment.

## What Genie 4 does

DeepMind says Genie 4 is natively multimodal: it generates audio together with the visuals, including ambient sound, music and character dialogue that reacts to what you do. You can change a scene with plain-language prompts, such as the weather, the camera, or added people and objects.

It is built on the Gemini family. DeepMind says Gemini Omni anchors how things look and behave, other Gemini models handle reasoning and continuity, and Gemini 3.8 Flash acts as a guiding agent that plans steps and checks progress during a simulation. The post also claims better physics, such as collisions and aerodynamics, and more natural human movement.

## Who can use it

Almost nobody yet. Google plans to release it first to Google AI Ultra subscribers inside Project Genie, then to early-access partners of the Gemini API and Google AI Studio. Interested users can join a waitlist. No pricing or general availability date was given.

## What DeepMind admits it hasn't solved

The post says visual details and object permanence can drift, especially in longer sessions, after several edits, or during navigation. It lists complex physical interactions, real-time audio accuracy, readable text, spatial accuracy and long-term consistency as open research areas.

It also leaves out basic numbers. The post does not state the output resolution or how long a session can run. Treat the "more realistic" claims as DeepMind's own until testers publish results.

All generated video and audio carry SynthID watermarks, and DeepMind says it ran internal safety evaluations before the preview.

## What to watch

The first sign of how good Genie 4 is will come from the trusted testers and from the Ultra rollout in Project Genie. Session length and resolution are the two numbers to look for.

## Sources

- [Genie 4: A new generation of world simulation](https://deepmind.google/blog/genie-4-a-new-generation-of-world-simulation/), Google DeepMind