Quick Answer
Genie 4 is Google DeepMind’s real-time interactive world model, announced October 27, 2026. It turns a text prompt (or a 16:9 reference image) into an explorable world you steer in plain language, with physics, ambient sound, music and characters that can talk back. It rolled out to eligible Google AI Ultra subscribers in Project Genie on November 18, 2026, and to developers via the Gemini API and Google AI Studio on November 19, 2026, at ~$0.10 per second of streamed output (model id models/genie-4, paid-tier key required). Google AI Pro members can only join a waitlist for now. It is still frontier research: API sessions last at most 120 seconds, and Google openly lists camera latency, intricate physical interactions, real-time audio precision and long-term scene permanence as unresolved problems.
What Is Genie 4?
A video model renders a fixed clip. A world model predicts the next frame from the current scene and your input, so the world keeps responding while you are inside it — closer to playing a game than watching a video. Genie 4 is Google’s most realistic and interactive world model yet, and it is natively multimodal: Gemini Omni grounds how the simulated world looks and behaves, and other Gemini models supply the reasoning that keeps it coherent.
The headline change over Genie 3 is that you can rewrite the world mid-session. In Genie 3 you started a world and lived with it. In Genie 4 you type a prompt while navigating — make it rain, add a character or an object, move the camera, redirect events — and the world changes without starting over. Sound is new too: Genie 3 worlds were silent, while Genie 4 generates native audio in real time — ambient sound, music and character dialogue — synchronised with the visuals.
Google also rebuilt the experience around it. The November 18 update brought a redesigned Project Genie — the browser-based world builder — with a conversational creation agent that helps you build the world, and a community gallery where you can remix other people’s worlds. Genie 3 stays in the model picker for longer runs and for worlds seeded from 360-degree panorama images.
Release Date and Availability
- October 27, 2026 — Google DeepMind announces Genie 4 as a research preview for trusted testers, naming Ultra subscribers in Project Genie and early-access Gemini API / AI Studio partners as the next release groups.
- November 18, 2026 — Genie 4 rolls out to eligible Google AI Ultra subscribers in the redesigned Project Genie. Google AI Pro members can join a waitlist on the same page; they are not part of the initial rollout.
- November 19, 2026 — Genie 4 opens to developers through the Gemini API (via the Interactions API) and Google AI Studio. Model id
models/genie-4, stable, with a paid-tier API key required. Serving regions are the United States, Western Europe and Japan.
Access rules: a personal Google Account, age 18 or over, in a country where Google AI Ultra is available. Project Genie is available in over 140 countries where Ultra is sold, according to Google’s plans page. Google’s plans page lists Ultra from $99.99 per month (pricing varies by country and promotion); AI Pro is $19.99 per month but its members can only join the Project Genie waitlist for now. The rollout is gradual, so not every eligible subscriber sees Genie 4 on day one.
Key Genie 4 Features
1. Worlds you can rewrite while you play
Type prompts mid-session to alter the weather, add a character or object, move the camera or redirect events — no restart needed. This is the feature Google leads with, and the one that separates Genie 4 from every video model.
2. Real-time native audio — a first for the series
Ambient sound, music and character dialogue are generated live with the visuals. Genie 3 had no audio of its own.
3. Higher fidelity visuals and physics
Google positions Genie 4 as its higher-fidelity option for visuals and physics, with more natural character movement and responsive real-world physics than Genie 3.
4. A creation agent and a remix gallery
The redesigned Project Genie adds a conversational agent that builds worlds with you — describe your vision in natural language and it orchestrates Google’s multimodal models to craft the look and a matching soundtrack — plus a community gallery for sharing and remixing worlds. Curated examples include “Living Memory” (turn personal photos into explorable spaces), “Dino Field Guide” (prehistoric habitats from 360-degree panoramas) and “Rock Paper Scissors” (Gemini as a real-time referee).
5. SynthID watermarking on everything
All video and audio generated in Project Genie carries SynthID, Google’s imperceptible watermark for identifying AI-generated content, and all inputs and outputs go through safety screenings.
6. Built on Gemini
Gemini Omni provides the real-world reference for how simulations look and behave; other Gemini models supply reasoning for coherence and continuity. One report names Gemini 3.8 Flash as the real-time guiding loop — treat that as reported, not confirmed by Google’s announcement.
Genie 4 Pricing Explained
There is no single price, because there are two very different doors in:
| Access route | Price |
|---|---|
| Project Genie (Google AI Ultra subscribers) | Included in Ultra — from $99.99/month on Google’s plans page; no per-second charge |
| Gemini API / Google AI Studio (developers) | ~$0.10 per second of streamed output (~$6.05 per minute); paid-tier API key required |
| Google AI Pro members | No access yet — waitlist only |
| Free tier / public demo | None — no free public version |
The API price is token-based, from Google’s Gemini API pricing page (paid tier, per 1M tokens):
| Item | Genie 4 price |
|---|---|
| Input tokens (text / reference image) | $1.50 per 1M tokens |
| Output tokens (text) | $9.00 per 1M tokens |
| Output tokens (video, audio included) | $17.50 per 1M tokens |
| Free tier | Not available |
| Batch API / Flex / Priority | Not supported |
| Context caching | Not supported |
Billing is based on emitted video output tokens at 240 tokens per frame — 5,760 tokens per second at 24fps — which is how Google gets to about $0.10 per second. A worked example: 60 seconds of streamed world costs about $6.05, and a full 120-second session costs about $12.10, plus a few cents of input for the prompt. Always call complete() to end a session, or billing keeps running.
Genie 4 vs Genie 3
| Spec | Genie 3 | Genie 4 |
|---|---|---|
| Announced | August 2025 | October 27, 2026 |
| Audio | None | Native audio — ambience, music, dialogue |
| Mid-session edits | No | Yes — weather, characters, camera, events |
| Session length | Up to 3 minutes per generation | Up to 120 seconds per API session |
| Output | 720p worlds | 720p at 24fps, 48 kHz stereo audio |
| Access | Project Genie (Ultra) | Project Genie (Ultra) + Gemini API / AI Studio |
| Best for | Longer runs, panorama-seeded worlds | Higher fidelity, sound, live rewriting |
Genie 3 is not retired — it stays selectable in Project Genie for the jobs it does better. Pick Genie 4 when the world needs sound and live changes; pick Genie 3 when you want the longest run or a world built from a 360-degree photo.
Performance and Benchmarks
All performance figures here are Google’s own claims from its launch posts and developer documentation, unless marked otherwise, and have not been independently verified by faqsnow.
There are no published benchmarks for Genie 4 — no scores, no leaderboards, no independent evaluation. That is normal for a world model: there is no standard test for “does the world stay coherent when you change the weather mid-scene”. The numbers Google does publish are engineering limits, not quality scores: 720p at 24fps, 48 kHz stereo audio, sessions up to 120 seconds with coherence strongest in the first 45–90 seconds, and serving from the US, Western Europe and Japan (higher latency elsewhere can reduce frame rate). Treat the article as the manufacturer’s spec sheet, not a review — the honest signal so far is that Google itself lists the model’s open problems in the same post as the launch.
How to Use Genie 4
1. If you have Google AI Ultra, open Project Genie
Eligible Ultra subscribers can try the new Project Genie experience now, in the browser. Sign in with a personal Google Account (18+). If you do not see Genie 4 yet, the rollout is gradual — check again rather than assuming you are ineligible.
2. If you only have Google AI Pro, join the waitlist
Pro members can join the waitlist on the Project Genie page. Google emails members as places open. There is no other access route for Pro today.
3. If you are a developer, use the Gemini API or AI Studio
From November 19, Genie 4 is available through the Gemini API (Interactions API, v1beta) and Google AI Studio with a paid-tier key, model id models/genie-4. Seed a world from a text prompt, a 16:9 reference image, or both, then steer with WASD or arrow keys and mid-session text events. Budget by the second — about $0.10 per streamed second — and keep first sessions short. Always call complete() when you finish, so billing stops.
4. Start with a simple world, then rewrite it live
Describe one clear scene first, walk through it, then try a mid-session change — rain, a new character, a camera move — to see what the model holds together and where it drifts. That is the fastest way to learn its limits.
5. Remember the watermark
Everything you generate carries SynthID. Do not present a Genie 4 world as real footage — it is marked as AI-generated by design.
Best Uses
- Prototyping interactive experiences — test a game level, a walkthrough or a story scene before building it for real (Google’s own first suggested use)
- Visualising spatial designs — walk through a room, a set or a layout from a text description or a reference image
- Training robotics and AI agents — generate interactive simulated environments to practise in (Google’s third suggested use)
- Exploring and remixing worlds — the Project Genie gallery, from “Living Memory” photo worlds to community remixes
- Creative experiments — the agent-built worlds and creator collaborations like “Villa Giallo” and “The Life of Ants” show what the tool is for right now: exploration, not finished production
Limitations and Things to Know
- Sessions are short. API sessions last at most 120 seconds, and Google says visual coherence and object permanence are strongest in the first 45–90 seconds. Longer runs drift — objects change shape or disappear, and scene permanence breaks down.
- Camera latency, intricate physical interactions and real-time audio precision are all listed by Google as unresolved challenges, along with long-term scene permanence.
- No free tier. The API needs a paid-tier key, and there is no free public demo. The cheapest way to try it is a short API session — a few seconds costs a few dozen cents.
- No Batch, Flex or Priority pricing, and no context caching. Every second is billed at the standard rate, and function calling, grounding, code execution and structured outputs are not supported.
- Serving regions are limited. The API serves from the United States, Western Europe and Japan; higher latency elsewhere can reduce frame rate.
- It is a research system, not a finished product. Google calls both Genie models frontier research systems. There are no independent benchmarks yet, and the demos are company-selected.
- No offline clips. For fixed-duration clip generation, multi-turn video editing or upsampling to 4K, Google points developers to Gemini Omni 1.2 Flash instead.
The honest read: Genie 4 is a glimpse of playable worlds, not a tool for finished work. Use it to prototype an idea, test a scene, or explore — not to produce a final video or a reliable training environment yet.
Frequently Asked Questions
What is Genie 4?
Google DeepMind’s real-time interactive world model. It turns a prompt into an explorable world with sound that you can steer and rewrite in plain language while you move through it.
When was Genie 4 released?
Announced October 27, 2026 as a research preview; rolled out to Google AI Ultra subscribers in Project Genie on November 18, 2026; opened to developers via the Gemini API and AI Studio on November 19, 2026.
How much does Genie 4 cost?
In Project Genie it is included in a Google AI Ultra subscription (from $99.99/month on Google’s plans page). For developers, the Gemini API costs about $0.10 per second of streamed output — $1.50 per 1M input tokens and $17.50 per 1M video output tokens (audio included). There is no free tier.
Can I use Genie 4 for free?
No. There is no free public version or demo. The routes in are an Ultra subscription, a paid-tier API key, or the waitlist if you are on AI Pro.
What is the difference between Genie 4 and Genie 3?
Genie 4 adds native audio and live mid-session editing, with higher visual fidelity — but shorter sessions (up to 120 seconds per API session). Genie 3 stays available for longer runs (up to 3 minutes) and panorama-seeded worlds.
How long can a Genie 4 session last?
Up to 120 seconds per API session, and Google says coherence is strongest in the first 45–90 seconds. Project Genie does not publish a Genie 4 session length.
Is Genie 4 available in my country?
Project Genie is available wherever Google AI Ultra is sold — over 140 countries, according to Google’s plans page — for eligible Ultra subscribers aged 18+. Developer access via the API serves the US, Western Europe and Japan.
What is the model id?
models/genie-4 (also accepts genie-4), stable, latest update November 2026.
Does Genie 4 make videos?
Not exactly. It makes playable worlds, not fixed clips. You can record what happens inside a world, but the output is an interactive simulation, and every clip carries a SynthID watermark.
What can I build with it?
Google names three uses: prototyping interactive experiences, visualising designs for spatial concepts, and training robotics and AI agents. For finished video clips, Google points to Gemini Omni 1.2 Flash instead.
Final Verdict
Genie 4 is the most important AI launch of the week that almost nobody can use yet — a world you can walk through, talk to and rewrite live, with sound, for about $0.10 a second. The honest limit is the session length: 120 seconds is a demo, not a game, and Google says so itself. But the direction is clear — video that you watch is becoming worlds that you enter — and the API price makes it cheap enough for any developer to try a two-minute world for about $12. If you have Ultra, open Project Genie; if you build, start with one short session and one live edit, and you will learn more in two minutes than any review can tell you.
Editorial update: November 20, 2026. This article is based on Google’s official announcements “Genie 4 is now available” (18 Nov 2026) and “Genie 4 now available through the Gemini API” (19 Nov 2026), Google DeepMind’s Genie 4 announcement (27 Oct 2026), and Google’s Gemini API documentation and pricing pages. Performance figures are Google’s own unless marked third-party, and have not been independently verified by faqsnow.