Google DeepMind · World Model · October 2026

Genie 4: Google's World Model You Can Talk To — and Steer While It Runs

Google DeepMind announced Genie 4 on October 27, 2026, and the headline is sound. Genie 4 is DeepMind's first real-time world model to generate audio natively: music, ambient noise, and interactive dialogue are generated simultaneously with the world itself, so the people in a scene can talk back. The second headline is control: you can change the weather, move the camera, or add a person mid-simulation, in plain language, while the world keeps running. The honest headline is the third one: this is a research preview for trusted testers, not a product you can buy today — so this page separates what DeepMind actually announced from what is still reported or unconfirmed.

Last updated: October 28, 2026. Facts checked against Google DeepMind's October 27, 2026 announcement; the two-minute duration figure is attributed to launch-day reporting, not to DeepMind's announcement text; price, resolution, and frame rate were not published at announcement and are marked unconfirmed below.

At a Glance

What Is a World Model, Exactly?

A video generator produces a clip. A world model produces a place. You start Genie 4 from a text prompt or an image, then move through the result with standard keyboard controls while the model simulates what happens next, frame by frame: how the world responds, how it sounds, and how the people in it move, react, and speak. DeepMind frames this as a step toward AGI — not because a playable beach scene is AGI, but because understanding and simulating the world is part of building systems that can plan and act in it, from robot training environments to interactive experiences.

That distinction decides whether Genie 4 is relevant to you. If you need a finished clip for a video, a video model like Veo 3.1 or Sora 2 is the right tool. If you need something you can step into, test, and change — a level sketch, a scene blocking, a training scenario — a world model is the category to follow, and Genie 4 is its newest public milestone.

The Four Advances, in Plain Terms

1. Real-time steering. While a world runs, type a change — new weather, a different camera angle, a new person or object — and Gemini adapts your prompt to fit the surrounding scene. Between your inputs, Gemini monitors and steers the scene so it keeps progressing cohesively instead of stalling or wandering.

2. Realism in physics and people. DeepMind says Genie 4 simulates real-world physics such as aerodynamics and collisions, and calls its simulated people the most dramatic leap: characters that move naturally, respond organically to their surroundings, and can be steered in real time.

3. Native audio. Every world carries the ambient sound of its own setting, from wind in the trees to waves at the beach — and the frontier, in DeepMind's words, is human speech. Characters converse and react to your inputs or to events in the world, all in real time. All demo audio in DeepMind's announcement was produced directly by Genie 4.

4. Multi-step events, guided by Gemini. Gemini 3.8 Flash serves as a guiding agent in the loop: it observes the scene, breaks goals into a sequence of steps, reviews incoming frames to verify each step, and adapts its plan when you intervene. DeepMind is also experimenting with Gemini 4 Argon and Genie 4 together, for example breaking down a humanoid robot task into steps that Genie renders in real time. That is research, not a shipping feature — but it shows where the category is headed: worlds with reliable causality, not just pretty motion.

Genie 4 vs Genie 3

How to Get Access (and What Not to Expect)

There is no public signup page that grants Genie 4 today, and no standalone price. The realistic path is: (1) join the waitlist for Genie 4 in Project Genie, linked from DeepMind's announcement — see the official announcement below; (2) if you are a Google AI Ultra subscriber, watch for the Project Genie rollout, which DeepMind describes only as coming soon; (3) if you build on the Gemini API or AI Studio, watch for early-access partner availability. DeepMind has published no general-availability date, so any third-party date you see is unconfirmed.

While you wait, the useful preparation is not a purchase — it is a test plan. Write down the one scene you would actually build first, what a success looks like (a camera move that stays coherent? a character that keeps its lines?), and what would make you stop (drift after 90 seconds? unreadable signage?). When access arrives, that list turns a novelty session into a real evaluation, and it protects you from building a project plan on a preview model's demo reel.

Official source: Google DeepMind — Genie 4: A new generation of world simulation (October 27, 2026). That page also carries the waitlist link for Genie 4 in Project Genie.

The Honest Caveats

Three things to keep straight. First, the demos are DeepMind's own recordings of real-time interactions — impressive, but not independent testing, and no reviewer had hands-on access at announcement. Second, the two-minute figure is widely reported but absent from DeepMind's announcement text, so treat it as a reported benchmark that could shift with the rollout. Third, everything about price, resolution, frame rate, and general availability is unconfirmed: if a page states those numbers for Genie 4 today without citing an official source, it is guessing. This page will be updated when DeepMind publishes them.

FAQ