Genie 4 API status
What Google has announced for developers, and what is still unknown.
Genie 4 vs Genie 3: what changed
Last updated:
Genie 3 came out in August 2025 and now powers Project Genie. Genie 4 was announced on October 27, 2026. The biggest changes Google describes are built-in sound and speech, changing the world with plain-language instructions while you are in it, and Gemini helping run multi-step scenes. Here is how the two compare, based only on what Google has published.
At a glance
- Genie 3 announced
- Aug 5, 2025
- Genie 4 announced
- Oct 27, 2026
- Genie 4 for everyone
- Not announced
Genie 4 vs Genie 3 side by side
Google has not published resolution or frame-rate numbers for Genie 4, so we leave that cell empty rather than copy Genie 3’s numbers across. This is a comparison of Google’s published descriptions, not our own test.
| Genie 3 | Genie 4 | |
|---|---|---|
| Made by | Google DeepMind | Google DeepMind |
| Announced | Aug 5, 2025 | Oct 27, 2026 |
| What it does | Real-time worlds from a text prompt | Real-time worlds with sound, speech and live language control |
| Published specs | 720p, 24 fps, consistent for a few minutes | Not published |
| Where you can use it | Project Genie (Ultra) | Coming to Project Genie |
| API | Not announced | Early Gemini API partners announced; no public date |
What’s new in Genie 4
Genie 3 generates real-time, explorable worlds: you type a prompt and move through the result at 720p and 24 frames per second, with the world staying consistent for a few minutes. It also introduced promptable world events, meaning you could change the scene with text while exploring. Genie 4 builds on that in four ways, according to DeepMind. First, it is natively multimodal: it generates ambient sound, and characters can talk and react to you in real time. Second, natural-language control is a headline feature, with examples like changing the weather, moving the camera or adding a new person or object. Third, DeepMind says physics and human movement look more realistic. Fourth, Gemini can watch the scene, break a goal into steps and guide what happens next. These are Google’s claims based on its own demos; there are no independent benchmarks yet.
Genie 3 in Project Genie is not the full research model
If you use Project Genie today, you are using Genie 3 together with Google’s Nano Banana Pro image model and Gemini. But the app does not include everything shown in DeepMind’s Genie 3 research. Promptable events, for example, were not included at launch. And while the research model stays consistent for a few minutes, the app launched with a 60-second limit per generation. Expect the same pattern with Genie 4: what reaches the app may differ from the demo video, so check the app’s own help pages once the rollout begins.
From announcement to app
Genie 3 took 177 days to go from announcement to a consumer app (our calculation from the two official dates). That is one data point, not a forecast for Genie 4.
- Aug 5, 2025
Genie 3 announced
Limited research preview.
- Jan 29, 2026
Project Genie launched
Built on Genie 3. Ultra subscribers in the US, 18+. More countries added since.
- Oct 27, 2026
Genie 4 announced
Trusted testers first, then Project Genie and early Gemini API partners.
- Not announced
Genie 4 for everyone
Google has not given a date.
Which one can you use?
Genie 3 is the one you can use today, through Project Genie, if you have a Google AI Ultra subscription, are 18 or older and live in a supported country. Genie 4 is limited to trusted testers for now. Google says it will come to Project Genie for Ultra subscribers and to early partners of the Gemini API and Google AI Studio, but without a date. Neither model has a public API you can sign up for and call today; when we last checked the Gemini API model list on October 28, 2026, there was no Genie entry. Keep in mind that an Ultra subscription gives you the Project Genie app, not API access.
Limits both models share
A newer version number does not mean the old problems are gone. DeepMind lists these Genie 4 limitations: visual details and object permanence can drift over time, real-world places are not reproduced accurately, readable text usually appears only when it is in your prompt, audio accuracy is still being worked on, and continuous interaction lasts a few minutes. Genie 3’s announcement also described consistency lasting only a few minutes. If what you want to build depends on objects staying put, exact text or long sessions, neither model is designed for that yet.
Get notified when the Genie 4 API opens.
One email when the API opens. Nothing before that.
No spam · Unsubscribe anytime Privacy
FAQ
What is the difference between Genie 4 and Genie 3?
Genie 4 adds native sound and speech, stronger natural-language control of the scene, more realistic physics and movement according to Google, and Gemini-guided multi-step simulation. Genie 3 powers Project Genie today.
Is Genie 4 available yet?
Only to trusted testers. Google says it is coming to Project Genie and early Gemini API partners next, with no public date.
What resolution does Genie 4 run at?
Google has not published it. Genie 3 was announced at 720p and 24 frames per second; we do not assume Genie 4 is the same.