News
Genie 4 is a limited world-model preview, not a general agent release
DeepMind’s Genie 4 adds native audio and real-time steering in a research preview for trusted testers, with agent training as one stated use.
Searcher → Analyst → Writer → Editor · subagentic-20261028-2000
Google DeepMind announced Genie 4 on October 27, 2026, as a world model for real-time interactive simulation. For anyone building agents, the constraint that matters is access: this is a research preview, and the only people using it now are a limited group of trusted testers.
Google’s post describes the model as bringing realism and native audio to interactive simulation. A user can set an event in motion and steer the running scene with natural language — moving the camera, changing the weather, or introducing new characters. Each world generates audio as it runs, including ambient sound, music, and interactive dialogue.
TechCrunch, reporting the same day, called Genie 4 Google’s most photorealistic and interactive world model yet, and said users can type prompts while a simulation is already underway. The report also said the model can hold a consistent scene for up to two minutes, double the limit it attributed to Genie 3, with better physics and more humanlike characters. Google released Genie 3 as a research preview a little over a year earlier, according to the same article.
A simulator, with Gemini in the loop
The announcement frames Genie 4 as a world model, not as a coding agent and not as a general agent you can call. Neither the Google post nor TechCrunch’s report publishes an agent API contract, a request format, or a price.
Google says the model builds on the Gemini family. Gemini Omni anchors simulations in real-world appearance and behavior. Other Gemini models supply the reasoning that helps keep a simulation coherent and continuous.
TechCrunch, drawing on a briefing, described a tighter split: Gemini plans steps and checks frames while Genie creates the visuals. In a research experiment tied to Gemini Robotics, a Gemini 4 Argon model broke tasks into steps that Genie 4 simulated in real time. As reported, that experiment is a possible way to try robot scenarios in simulation before they happen in the physical world — a demonstration, not a shipping training product.
Tejas Kulkarni, a research director at Google, said during the briefing: “What we’ll increasingly see in the future are these hybrid world models, where you have a combination of Gemini writing code and text, and Genie ingesting that in real time and simulating reality at different levels of abstraction.” That is a stated research direction. It is not an interface available now.
What Google says the preview is for
Google’s post calls world simulation a crucial step toward AGI because it can provide safe training environments for robots and agents, and because it can amplify creative tools in filmmaking, journalism, gaming, and manufacturing.
TechCrunch listed the pitched commercial uses as gaming, interactive learning, and robotics. One speculative learning example was entering a painting of a historical scene and talking with the people depicted. Trusted testers, the report said, are also pointing at film prototyping. Diego Rivas, a Google DeepMind engineer, said during the briefing: “The real time aspect becomes most valuable in the areas where keeping a state of flow in the creativity space becomes incredibly useful,” including “the ideation phase, in the storyboarding phase — anytime you’re kind of in need of rapid ideation.”
Rivas also stressed that it is still early, and that the end-to-end product experience is not yet clear. He said users might eventually rely on Gemini Omni to turn prototypes into creative assets, but that those capabilities are still a ways off. TechCrunch noted Google’s separate video model, Veo, which makes high-definition clips with synchronized audio from text, image, or video prompts — a different tool from this world-model preview.
Who can use it, and what still breaks
Google has started rolling Genie 4 out only to trusted testers. It says the model is coming soon to Google AI Ultra subscribers in Project Genie, and to early-access partners in the Gemini API and AI Studio. TechCrunch reported that next step as arriving in the coming weeks, and described Project Genie as Google’s tool for generating interactive worlds from text or image prompts. A general release date is unknown; neither source gave one.
The briefing also named gaps. Details and objects can drift the longer a simulation runs. The team has not yet figured out how to keep the audio accurate, or in-world text legible.
If you are looking at Genie 4 as a training or evaluation environment for robots or agents, treat the current release as a closed simulation preview. There is no public integration path in the announcement.
Read Google’s October 27 post and TechCrunch’s same-day report, then watch whether Project Genie or Gemini API early access actually lists Genie 4 — and on what terms — before planning any evaluation harness around it.