Genie 4 vs Genie 3: What Changed
Genie 4 is Google DeepMind's first real-time world model to generate audio natively, and it uses Gemini to interpret prompts and plan inside the world. Both models let you explore a generated world in real time and change it with text for a few minutes at a time. Below is everything Google has published about the two, side by side, and what it has not said yet.
Genie 3 vs Genie 4 at a Glance
Only facts Google DeepMind has published. "Not published" means Google has given no figure, not that the feature is missing.
| Genie 3 | Genie 4 | |
|---|---|---|
| Announced | August 5, 2025 | October 27, 2026 |
| What it is | Real-time world model | Real-time world model that builds on Gemini: Gemini Omni for real-world appearance and behavior, other Gemini models for reasoning |
| Explore in real time | Yes, with navigation controls | Yes, with keyboard, mouse and text prompts |
| Native audio | Not mentioned by Google | Yes: ambient sound, music and characters that speak |
| Change the world with text | Yes, via promptable world events | Yes, live, with Gemini interpreting prompts and planning |
| Physics and people | Not compared by Google | Described as a new standard of realism for real-time simulation |
| Session length | A few minutes, visual memory up to about one minute back | A few minutes of continuous interaction |
| Resolution and frame rate | 720p at 24 fps | Not published |
| Watermarking | Not stated on the Genie 3 page | SynthID on all generated video and audio |
| Access at launch | Limited research preview for a small group of academics and creators | Research preview for trusted testers |
| Next access step | Project Genie on Google Labs, linked from the Genie 3 page | Google AI Ultra in Project Genie, then Gemini API and Google AI Studio early access partners |
| Public price | Not published | Not published |
What's New in Genie 4
The upgrades Google DeepMind highlights in the Genie 4 announcement.
Sound in the same loop
Google calls Genie 4 its first real-time world model to generate audio natively. Leaves crunch underfoot, crowds cheer, and characters can talk with you and with each other.
Gemini as a planner
Gemini works behind the scenes to interpret prompts and make plans inside a changing world. In a robotics demo, Gemini writes multi-step plans that Genie 4 receives as text captions.
More realistic motion
Google says Genie 4 sets a new standard of realism for real-time interactive simulation, including physics and human movement. It does not publish a benchmark against Genie 3.
What Stays the Same
Some things Genie 3 already did, and some limits that carry over.
Short sessions
Both models are described as supporting a few minutes of continuous interaction, not hours. Genie 4 is not a longer-running version of Genie 3.
Text-driven changes
Genie 3 introduced promptable world events, such as changing the weather or adding objects and characters. Genie 4 keeps this and adds Gemini to interpret and plan.
On-screen text, and drift in Genie 4
Both models usually render legible text only when it is written into the world description. Google lists drift in details and object permanence as a Genie 4 limitation; for Genie 3 it says worlds stay largely consistent for several minutes.
Which One Can You Use?
Neither is generally available. Here is the practical difference today.
If you want to try a Genie model
Project Genie on Google Labs is Google's public route. Genie 4 is planned to arrive there first for Google AI Ultra subscribers. Check the official model page for current eligibility.
If you want Genie 4 specifically
Join Google DeepMind's official Genie 4 waitlist. Signing up does not guarantee access, and Google has not given a date for wider release.
If you are a developer
Google plans Gemini API and Google AI Studio early access for partners. No public endpoint, model ID or price exists yet. Genie4.io does not offer an API.
Genie 4 vs Genie 3: Questions
Short answers, as of October 2026.
Is Genie 4 better than Genie 3?
Google presents Genie 4 as a step up: native audio, Gemini-based planning and more realistic physics and people. It has not published head-to-head benchmarks, so there is no measured score comparing the two.
Does Genie 4 last longer than Genie 3?
Not according to Google's published wording. Both are described as supporting a few minutes of continuous interaction. Google has not given a specific session limit for Genie 4.
Did Genie 3 have sound?
Google's Genie 3 announcement does not mention audio. Google describes Genie 4 as its first real-time world model to generate audio natively.
What resolution does Genie 4 run at?
Google has not published a resolution or frame rate for Genie 4. Genie 3 was announced at 720p and 24 frames per second.
Can I still use Genie 3?
Genie 3 launched as a limited research preview, and Google has pointed people to Project Genie on Google Labs. Availability depends on Google's current rollout and your region.
Is this page official?
No. Genie4.io is an independent site and is not affiliated with Google or Google DeepMind. Genie is a trademark of Google LLC. Every fact here links back to Google DeepMind's own pages.
Sources
Google DeepMind's official pages used for this comparison.
- Genie 4: a new generation of world simulation ↗
Google DeepMind blog, October 27, 2026.
- Genie 3: a new frontier for world models ↗
Google DeepMind blog, August 5, 2025.
- Genie model page ↗
Official overview, limitations, SynthID and the Genie 4 waitlist link.