Google DeepMind Unveils Genie 4, a World Model With Native Audio and Real-Time Steering

Google DeepMind on October 27, 2026 announced Genie 4, a world model that generates interactive environments together with their own sound and spoken dialogue, and that users can steer with text while a simulation runs. It is available first as a research preview for trusted testers, with Google AI Ultra subscribers and early access developers next in line, the company said.

A world model generates a space frame by frame as someone moves through it, instead of rendering a finished clip. Genie 4 arrives two weeks after Google shipped Gemini 4 Argon, and it leans on the Gemini family more than any earlier Genie: Gemini Omni anchors how scenes look and behave, and other Gemini models plan the action. It is the latest of Google's model releases aimed at simulation for robots and agents as much as at entertainment.

What is new in Genie 4?

Genie 4 adds native audio, live text steering and more realistic physics and people, Google DeepMind said in its October 27 announcement. It calls Genie 4 its first real-time world model to generate audio natively: each world carries ambient sound such as wind or waves, and characters can hold dialogue that reacts to the user's input or to events in the scene.

Users move with standard keyboard controls and can type prompts mid simulation to change the weather, move the camera or add a person or object. Gemini rewrites each prompt to fit the running scene and keeps steering the world between inputs, DeepMind says. The company also claims better simulation of aerodynamics and collisions, and calls the way Genie 4 renders people, with more natural movement and reactions, its "most dramatic leap."

Scenes now hold together for about two minutes, up from one minute for Genie 3, The Verge and TechCrunch report. DeepMind's own post gives no session length, resolution or frame rate.

How does Gemini work inside Genie 4?

Gemini 3.8 Flash acts as a guiding agent in the loop, according to Google DeepMind: while Genie 4 renders the world, Gemini watches the scene, splits a goal into steps and checks incoming frames to confirm each step worked. When the user intervenes or the world shifts, it adjusts the plan.

DeepMind also showed a research experiment that links Genie 4 with Gemini 4 Argon and its Gemini Robotics work. Argon broke a task into steps that were fed to Genie 4 as text captions, producing a simulation of Apptronik's Apollo humanoid robot playing checkers, with no physical robot involved. DeepMind describes this pairing as an early step toward "hybrid world models" that combine the reasoning of language and code models with pixel level simulation.

How does Genie 4 compare with Genie 3?

Genie 4 roughly doubles how long a world stays coherent and adds sound, which Genie 3 lacked, but Google has published fewer technical specs this time. When DeepMind announced Genie 3 on August 5, 2025, it listed 720p resolution at 24 frames per second and said worlds stayed consistent for "a few minutes." The Genie 4 post gives neither a resolution nor a frame rate.

The doubling The Verge and TechCrunch report is measured against the product, not the research claim. When Google opened Project Genie, its Genie 3 based web app, on January 29, 2026, it capped generations at 60 seconds, so two minutes is twice that cap rather than twice the "few minutes" DeepMind described in 2025.

Genie 4 also lands in a crowded field. Runway released GWM Worlds 2, a playable world model that also generates audio, a couple of months earlier, and AMD is acquiring Fei-Fei Li's World Labs, according to TechCrunch.

Who can use Genie 4, and when?

Only trusted testers can use Genie 4 as of October 27, 2026; Google says it is "coming soon" to Google AI Ultra subscribers in Project Genie and to early access partners of the Gemini API and Google AI Studio, and it has opened a waitlist. No date has been given. Ultra starts at $99.99 per month in the US, according to Google's plan page, and Ars Technica reports Google may later extend access to its cheaper AI Pro and AI Plus plans.

DeepMind says testers in creative industries, journalism, education, gaming and engineering have used Genie 4 to run quick "takes" of a scene, turn archival photos into explorable spaces and prototype game mechanics before writing engine code. All video and audio it generates will carry SynthID, Google's invisible watermark for AI content. The company lists drift over longer sessions, real-time audio accuracy, legible text, spatial accuracy and long-term consistency as unsolved.

What we don't know yet

• When Ultra subscribers and API partners will get access, and what developers will pay.
• Whether Genie 4 will reach users outside the US, where Project Genie launched.
• Genie 4's resolution and frame rate, and how it performs in independent tests.
• How DeepMind's robotics experiments with Genie 4 will be used beyond the demos shown.