CONFIRMED
Google DeepMind confirmed on Tuesday, October 27, 2026, the launch of Genie 4, the new generation of its world model, the technology that generates interactive virtual environments frame by frame, without needing to program a scene or pre render anything. The announcement went out on Google's official blog and was detailed in a press briefing with DeepMind researchers, with independent coverage from TechCrunch, Ars Technica and Bloomberg confirming the same figures and statements.
What Genie 4 does differently
The big shift from Genie 3, released just over a year ago, is native audio: every generated world now comes with ambient sound, music and interactive dialogue born alongside the image, not added afterward. Users type a command and change the scene in real time, swapping the weather, moving the camera or introducing a new character into the story. Under the hood, Gemini Omni anchors the world's appearance and behavior, while Gemini 3.8 Flash acts as a scene director, checking frame by frame whether the simulation still makes sense. The result can hold a coherent scene for up to two minutes, double the previous version's limit.
Why this matters beyond developers
A world model is not just a visual toy. DeepMind describes Genie 4 as part of a broader path toward artificial general intelligence, because training robots and agents in realistic simulated environments is far cheaper and safer than testing directly in the physical world. In an internal experiment with Gemini Robotics, an action model broke a task into steps that Genie 4 simulated in real time before any real execution. Outside robotics, the targeted commercial uses are film, games and journalism: writers and game designers are already using Genie 4 to prototype scenes and levels without drawing or rendering anything by hand, just describing what they want to see.
Access is still restricted, and expensive
For now Genie 4 is in research preview, available only to a small group of trusted testers from the gaming and film industries. Google promises to widen access in the coming weeks, first to Google AI Ultra subscribers, a plan starting around one hundred dollars a month, inside the Project Genie tool, and then to early access partners through the Gemini API and Google AI Studio. There is still no set usage price for developers, and there is a waitlist for those wanting to join the preview queue.
The limit Google itself admits
DeepMind is direct about the limitations: Genie 4 still forgets how things started after a few minutes, and details and objects can warp or disappear in longer simulations. The team itself treats the project as ongoing research, not a finished product, and stresses that all generated audio and video carries the SynthID watermark, Google's system for flagging AI generated content. It is a real launch, but one the company itself treats as an advanced prototype, not a general purpose tool ready for everyone.
Sources
Google: We're announcing Genie 4, our new world model; TechCrunch: Google's Genie 4 world model adds audio, more realistic worlds, and real time steering; Ars Technica: DeepMind's Genie 4 world model adds audio and more realistic people to simulations; Quartz: Google DeepMind is unveiling Genie 4, its most realistic AI world model yet.
