# What Is Genie 4? Google DeepMind’s New World Model, Explained URL: https://whatisagenticai.net/blog/what-is-genie-4/ Published: 2026-10-27 Publisher: Little Tough LLC, maker of AI Agent Assistant: Agentic AI (https://apps.apple.com/us/app/ai-agent-assistant-agentic-ai/id6761501260) Genie 4 is Google DeepMind’s new world model, announced October 27, 2026. It builds playable worlds with sound in real time. Who can use it and what is new. Google DeepMind announced Genie 4 on October 27, 2026, and the name is suddenly all over tech news and gaming forums. If you saw "Genie 4 AI" in a headline and wondered whether it is a chatbot, a video maker, or a game engine, the honest answer is: a bit of all three, and none of them exactly. This page explains what Genie 4 is, what is new compared with Genie 3, who can use it right now, and why it matters for AI agents, not just for games. **TL;DR:** Genie 4 is Google DeepMind’s newest "world model": AI that turns a text or image prompt into a 3D world you can walk around in with a keyboard and mouse, generated live as you move. New in Genie 4: it makes its own sound (music, ambient noise, characters talking), you can change the scene mid-play by typing, and a scene stays consistent for up to about two minutes, double Genie 3. It is a research preview for trusted testers today, coming "soon" to Google AI Ultra subscribers in Project Genie. No price has been announced. ## What is Genie 4? Genie 4 (also written "Google Genie 4" or "DeepMind Genie 4") is an AI model from Google DeepMind that simulates a world in real time. You describe a place, like "a rainy market street at night," and Genie 4 draws it frame by frame while you steer through it. Turn left, and it invents what is to your left. Walk into a shop, and it invents the shop. That is the key difference from a video model. A video model like Google’s Veo makes a fixed clip you watch. A world model reacts to what you do, so every run is different. DeepMind calls Genie 4 "our most realistic and interactive world model yet." Under the hood it is not working alone. According to DeepMind, Gemini models read your prompts and make plans as the world changes, and Gemini Omni keeps the look and behavior of things close to the real world. TechCrunch describes the setup as Gemini acting like a director that plans steps and checks frames, while Genie paints the visuals. ## What is new in Genie 4 vs Genie 3? Genie 3 came out as a research preview in August 2025. Genie 4 is a little over a year newer, and the changes are easy to see. | | Genie 3 | Genie 4 | | --- | --- | --- | | Announced | August 2025 | October 27, 2026 | | Sound | No audio | Native audio: music, ambient noise, footsteps, characters speaking | | Consistent scene | About 1 minute | Up to about 2 minutes (per TechCrunch) | | Change the world mid-play | Text "world events" | Free-form prompts: weather, camera angle, new people or objects | | Look and physics | Good | More photorealistic, better physics, more humanlike characters | | Access | Project Genie for Google AI Ultra (since January 2026) | Trusted testers now; Ultra and API partners "in the coming weeks" | Sound is the headline. DeepMind says the worlds come with "natively generated" audio, like leaves crunching or a crowd cheering, and TechCrunch reports characters can hold interactive dialogue. That is the first time a Genie model has done this. ## Genie 4 release date and who can use it Genie 4 was announced on Tuesday, October 27, 2026, as a **research preview**. Here is the rollout as Google describes it: - **Now:** a small group of trusted testers. - **In the coming weeks:** Google AI Ultra subscribers inside **Project Genie**, Google’s app for making interactive worlds from text or images. - **Also coming:** early access partners of the Gemini API and Google AI Studio. DeepMind’s Genie page links a waitlist form. There is no public date for everyone else yet. ## Is Genie 4 free? No, not right now. The only consumer path Google has named is Project Genie for Google AI Ultra subscribers, which is Google’s top paid AI plan. Google has not announced separate pricing for Genie 4 or for API access, and it has not said which countries get it first. We will update this page when those details land. ## Why Genie 4 matters for AI agents Most coverage talks about games, and Bloomberg pointed out that an earlier Genie release rattled video game stocks. But DeepMind’s own pitch is just as much about agents and robots. A world model is a practice ground. An AI agent can try a task thousands of times inside a simulated world without breaking anything real. DeepMind says Genie worlds are places where agents like its SIMA game agent "can learn new skills," and where its Gemini Robotics team can test robot plans before they happen in a real room. In one example on DeepMind’s page, a planning model named [Gemini 4 Argon](https://whatisagenticai.net/blog/what-is-gemini-4-argon/) breaks a robot job into steps and writes them out as captions, while Genie 4 shows each step happening live. That is the same plan, act, check loop we describe in [how AI agents work](https://whatisagenticai.net/blog/how-ai-agents-work/), just with a simulated world as the place where the acting happens. ## Genie 4 limitations Google is unusually direct about what does not work yet. From DeepMind’s page and TechCrunch: - Details and objects can drift the longer you play, after several edits, or while moving around. - Sessions last a few minutes, not hours. - Text inside the world is usually readable only if you wrote it into the prompt. - Real places cannot be copied with perfect accuracy. - Audio accuracy is still being improved. - Scenes with many people or agents interacting are still a research problem. Every video and audio frame Genie 4 makes carries Google’s SynthID watermark, so it can be identified as AI-made. ## Genie 4 vs other world models Google is not alone. TechCrunch notes that Runway released GWM Worlds 2 a couple of months ago, AMD is buying the world model startup World Labs, and Amazon is reportedly in talks to buy Decart. If you have followed AI video tools like [Kling 4.0](https://whatisagenticai.net/blog/what-is-kling-4-0/), think of world models as the next step: not a clip you watch, but a place you can step into. ## What this means for you Unless you are a trusted tester or pay for Google AI Ultra, you cannot use Genie 4 today. And even when it arrives, it is a research toy for a few minutes of play, not something that gets your week in order. The bigger story is the direction. Labs are building AI that does not just answer, it acts: in simulated worlds today, in real software and real tasks already. That is what [agentic AI](https://whatisagenticai.net/what-is-agentic-ai/) means: you hand over a goal, and agents plan the steps and do the work. You do not need a waitlist for that part. Our app, **Agentic AI**, gives you a team of ready-made expert agents on your iPhone and iPad, running on frontier models. Put up to 4 agents on one job, like researching a trip or drafting a project plan, with no account to create. Your first 3 tasks are free. ## Quick answers **What is Genie 4 AI?** A world model from Google DeepMind that turns a text or image prompt into an explorable 3D world with sound, generated in real time as you move through it. **When is Genie 4 coming out?** It was announced October 27, 2026 as a research preview for trusted testers. Google says it reaches Google AI Ultra subscribers in Project Genie and Gemini API early access partners "in the coming weeks." **Is Genie 4 free?** No. The announced consumer path is Project Genie, which needs a paid Google AI Ultra plan. No separate Genie 4 price has been announced. **Can I download Genie 4?** No. It runs on Google’s servers through Project Genie and, later, the Gemini API. There is no download. **Is Genie 4 a game engine?** Not really. It has no code or level files. It draws each frame from what it has learned, which is why scenes can drift after a couple of minutes. **Sources:** [Google DeepMind: Genie 4](https://deepmind.google/models/genie/) · [TechCrunch](https://techcrunch.com/2026/10/27/googles-genie-4-world-model-adds-audio-more-realistic-worlds-and-real-time-steering/) · [Unite.AI](https://www.unite.ai/google-deepmind-rolls-out-genie-4-world-model-in-research-preview/) · [The Decoder](https://the-decoder.com/google-deepminds-genie-4-puts-a-modern-spin-on-agent-designs-that-predate-llms/) · [TechCrunch on Project Genie](https://techcrunch.com/2026/01/29/i-built-marshmallow-castles-in-googles-new-ai-world-generator-project-genie)