Google DeepMind released Genie 4, its newest world model, in research preview on Tuesday, October 27, 2026 — and for the first time, the worlds it generates have sound, music, and people who talk back. Why it matters: world models are moving from silent video clips toward interactive simulations you can steer live, which is the direction robotics training, games, and film prototyping are all heading.
October 27, 2026
What happened
Genie 4 turns a text prompt or an image into an interactive world you move through in real time. DeepMind describes it as natively multimodal: the audio is generated alongside the scene, not added afterward — ambient noise, music, and interactive dialogue from characters inside the world. Trusted testers get access first. In the coming weeks, DeepMind says, it will roll out to Google AI Ultra subscribers in Project Genie, and to early access partners of the Gemini API and Google AI Studio.
What is new compared with Genie 3
Genie 3, previewed a little over a year ago, produced silent clips capped at about 60 seconds. Genie 4 doubles the consistent-scene limit to about two minutes and adds four main advances: native real-time audio, more photorealistic environments, better physics — including aerodynamics, collisions, and characters who move more naturally — and live steering. You can type a change while a simulation is running, such as new weather, a different camera angle, or an extra person or object, and the model adapts the scene around it. Set an event in motion, and Genie 4 simulates what happens next.
Gemini is wired into the loop more closely than before. DeepMind says Gemini Omni anchors the simulations in real-world appearance and behavior, while other Gemini models supply reasoning to keep a scene coherent. TechCrunch also reports that, in a Gemini Robotics experiment, a Gemini 4 Argon model broke tasks into steps that Genie 4 then simulated in real time — a possible way to rehearse robot scenarios in simulation before trying them in the real world.
The honest limits
DeepMind is candid that this is early. Visual details and object permanence can still drift the longer a simulation runs, especially after multiple complex edits, and the team told reporters it has not yet solved accurate audio in every case or legible in-world text. Access is also narrow for now: a research preview, not a public product.
Why it matters
The competitive context is moving fast — Runway released its second world model, GWM Worlds 2, a couple of months ago, and the field is consolidating around simulation for robots and self-driving. Genie 4’s real step is the combination: a world you can walk through, hear, talk to, and rewrite mid-scene, with a reasoning model directing it. If that holds up outside demos, the most valuable use may not be games at all, but cheap, safe, endlessly variable training worlds for agents and robots. Everything AI, tested — this one is worth watching as Ultra subscribers get their hands on it.
Frequently asked questions
Can I use Genie 4 now?
Only if you are one of DeepMind’s trusted testers. DeepMind says Google AI Ultra subscribers get it in Project Genie in the coming weeks, with Gemini API and AI Studio early access partners also included.
What is a world model?
Instead of rendering pre-built 3D assets, a world model generates the environment frame by frame in real time as you move through it, predicting how the scene should look, sound, and respond.
How is Genie 4 different from a video generator like Veo?
Veo makes a clip you watch. Genie 4 makes a world you control — you navigate it, change it by typing, and hear characters react while it runs.
Sources: TechCrunch, Google DeepMind (via Unite.AI and AI Weekly reporting on the DeepMind announcement)