Genie 4: Google DeepMind's world model adds audio and live steering, but it's trusted testers only, with no price yet

Genie 4 is Google DeepMind's newest world model, announced on October 27, 2026: a system that generates a scene you can walk through in real time, now with its own sound, characters that talk, and the ability to change the world mid-session by typing. Google says its Gemini models plan and check what Genie draws. You can't use it yet unless you're a trusted tester. Google says it's "coming soon" to Google AI Ultra subscribers in Project Genie and to early-access partners in the Gemini API, with a waitlist open now. World models are a crowded bet in 2026, as our story on Amazon's reported talks to buy Decart explains.

This story is based on Google DeepMind's October 27 blog post and Genie model page, the Genie 4 waitlist form, Google's posts on Project Genie from January 29 and May 19, 2026 and on AI subscriptions from May 19, the August 2025 Genie 3 post, and TechCrunch's October 27 report, checked October 28 and re-checked October 29, 2026.

The short version

• What it is: a model that generates an interactive world from a prompt and keeps it running as you move through it with keyboard controls and text prompts.
• New in Genie 4, per Google: native audio (music, ambient sound and dialogue), text-prompt edits in real time, more realistic physics and human motion, and Gemini 3.8 Flash acting as a planning agent.
• Who can use it: trusted testers only, as of October 29, 2026.
• Coming: Google AI Ultra subscribers in Project Genie, and early-access partners in the Gemini API and Google AI Studio. No date beyond "soon"; TechCrunch reported "in the coming weeks."
• Cost: no price for Genie 4 itself. Project Genie is part of the AI Ultra tier that Google listed at $200 a month in May 2026.
• Not published: resolution, frame rate, a technical report or a model card.

What is Genie 4?

Genie 4 generates an explorable world that responds as you interact with it. DeepMind's post, by Raia Hadsell, Diego Rivas and Hang Qi, says it "unlocks unprecedented realism" and calls it "our first real-time world model to generate audio natively." Characters can speak and react to what you do.

You steer it two ways. Standard keyboard controls move you and the camera. Text prompts change the scene while it runs: Google's examples include changing the weather, moving the camera, or adding people and objects. Google says Gemini adapts those prompts to the scene and keeps the world moving between inputs.

Google also describes Gemini models working alongside Genie. Gemini 3.8 Flash acts as a guiding agent: it watches the scene, breaks a goal into steps, checks frames to see whether a step is done, and revises its plan when the world changes or the user intervenes. Google says Gemini Omni helps anchor simulations in real-world appearance. In one experiment, Google paired Gemini 4 Argon, its frontier model (covered in our Argon launch story), with Genie 4 and Gemini Robotics to simulate a humanoid robot playing checkers.

Who can use Genie 4, and is there a waitlist?

Only trusted testers can use it today. Google's model page says Genie 4 is "available as a research preview for trusted testers," and the October 27 post says it is coming to:

• Google AI Ultra subscribers, in Project Genie, Google's existing web app for generating and exploring worlds;
• early-access partners in the Gemini API and Google AI Studio. Google doesn't mention general API availability.

The post links to a Google Form titled "Apply for Genie 4 access in Project Genie." The form doesn't show who qualifies, and Google hasn't said how applicants are chosen.

On countries and age, Google hasn't said anything specific to Genie 4. Project Genie's existing rules are the best guide: when it launched on January 29, 2026, it was for US AI Ultra subscribers aged 18 and over, and not for AI Ultra for Business users. On May 19, 2026, Google said it was "expanding access to Google AI Ultra subscribers globally," still for people 18 and older. Whether Genie 4 follows the same rules is not yet confirmed.

How much does Genie 4 cost?

Google hasn't announced a price for Genie 4. Access through Project Genie comes with the top Google AI Ultra plan. Google's May 19, 2026 subscriptions post lists "Project Genie (AI Ultra $200; global)," after cutting that tier from $250 to $200 a month and adding a cheaper $100 Ultra tier. It's not clear whether the $100 tier will get Genie 4; Google's October 27 post says only "Google AI Ultra subscribers."

There's no Gemini API price either; API access is limited to early-access partners. For how Google's consumer plans compare more broadly, see our Claude, ChatGPT and Gemini plans comparison.

What resolution and how long are Genie 4 worlds?

Google hasn't published a resolution or frame rate for Genie 4. On length, its model page says the model "can support a few minutes of continuous interaction, rather than extended sessions," and lists that as a limitation. TechCrunch reported on October 27 that Genie 4 keeps a scene consistent for "up to two minutes," which it called double its predecessor's limit. That figure doesn't appear on Google's own pages.

The comparison depends on which predecessor you mean. Genie 3, announced on August 5, 2025, ran at 720p and 24 frames per second and stayed "largely consistent for several minutes," according to DeepMind. But Project Genie, the consumer app built on Genie 3, capped generations at 60 seconds, per Google's January post, which also said Genie 3's "promptable events" for changing a world mid-session weren't included yet. Two minutes would be double that cap.

Last verified: October 29, 2026

What does DeepMind claim over Genie 3?

DeepMind's post makes qualitative claims, not measured ones. It says Genie 4 is more realistic, models physics such as aerodynamics and collisions, produces characters who move naturally and can be steered, generates audio natively, and accepts freeform edits in real time. It doesn't publish benchmarks, side-by-side numbers or a technical report, and it doesn't directly compare Genie 4 with Genie 3 in its text. Without those, there's no way to check how large the improvements are.

What are Genie 4's limitations?

Google lists them itself. The post says that while Genie 4 "can maintain coherent environments over short sessions, visual details and object permanence can still shift," especially in longer simulations, after several complex prompt edits, or while moving through a scene. The model page lists six limitations:

• keeping environments coherent over time;
• interactions with other agents;
• accuracy for real-world locations;
• rendering legible text;
• limited interaction duration;
• audio accuracy.

Rivas told TechCrunch it's "still early days" and that turning prototypes into finished creative assets is "still a ways off."

What has Google said about Genie 4 safety?

Google says Genie 4 "includes strict safety guardrails that are consistent with our safety policies across other models." It says it ran internal safety evaluations before the preview, released it first to trusted testers, and that all generated video and audio will carry SynthID, Google's invisible watermark. Google hasn't published the evaluations, a model card or a safety report for Genie 4, so these are Google's statements, not documented results.

What's confirmed, reported and uncertain

• Confirmed (Google): the October 27 announcement and authors; native audio, dialogue and real-time text edits; Gemini 3.8 Flash as planning agent; trusted-tester preview; planned access for AI Ultra in Project Genie and Gemini API partners; the waitlist form; the listed limitations; SynthID; Project Genie's $200 Ultra tier and 18+ rule as of May 19, 2026.
• Reported: the two-minute consistency figure and "coming weeks" timing (TechCrunch).
• Uncertain: resolution and frame rate; any price for Genie 4 or its API; whether the $100 Ultra tier is included; country and age rules for Genie 4 specifically; how testers are chosen.