DeepMind announces Genie 4, a real-time world model with native audio
DeepMind announced Genie 4 on 27 October 2026. The world model generates audio and dialogue as it simulates, and only trusted testers can try it.
The AI Pin
own
Google DeepMind announced Genie 4, a real-time world model with native audio, on 27 October 2026. Genie 4 is a research preview for trusted testers, not a general release. [1]
What did DeepMind announce?
DeepMind says Genie 4 "unlocks unprecedented realism and improves real-time interactive simulation". It is the lab's first real-time world model to generate audio natively, including music, ambient noise and interactive dialogue. Characters converse and react to user inputs in real time. The blog is by Raia Hadsell, Diego Rivas and Hang Qi. DeepMind posted the announcement on X at 16:10 UTC. [1] [2]
The daily pin – 5 minutes, sourced.
Five sourced stories each morning. Free, and you can unsubscribe anytime.
What can a user change?
Users navigate with standard keyboard controls and steer with text prompts. Natural language can change the weather, move the camera, or add a person or object in real time. Gemini adapts the prompt to the scene and steers between inputs. [1]
DeepMind says the model covers physics such as aerodynamics and collisions, and human action. In its words, "the most dramatic leap is how Genie 4 simulates people". [1]
Gemini Omni anchors simulations in real-world appearance and behavior. Other Gemini models provide reasoning for coherence and continuity. Gemini 3.8 Flash acts as a guiding agent in the loop, observing the scene, breaking goals into steps and verifying each step. [1]
Who can try Genie 4?
The preview is for trusted testers now. It is coming to Google AI Ultra subscribers in Project Genie, and to early-access partners of the Gemini API and Google AI Studio. A waitlist is open in Project Genie. It is not generally available. All Genie 4 video and audio will carry SynthID watermarking. [1]
What does DeepMind say it is for?
DeepMind frames world simulation as "a crucial aspect of building AGI". Uses include training environments for robots and agents, and new interactive experiences. The predecessor is Genie 3, from August 2025. [1]
A separate story on Gemini agents in Minecraft is at science/gemini-4-argon-agents-beat-minecraft-in-27-hours.
Where does DeepMind say it falls short?
DeepMind says visual details and object permanence can drift over longer runs. That can also happen after multiple complex prompt edits, or while navigating. DeepMind says Genie 4 stays coherent over short sessions. Complex physical interactions, real-time audio accuracy, legible text, spatial accuracy and long-term consistency remain open research areas. [1]
Tags: deepmind · genie · models · world-model
Why it matters
- Why it matters
- The world model generates audio and dialogue as it simulates, and only trusted testers can try it.
- What's confirmed
- Confirmed by primary source: Google DeepMind, Genie 4: A new generation of world simulation, 27 October 2026, by Raia Hadsell, Diego Rivas and Hang Qi. The X post is timestamped 16:10 UTC the same day.
- What's uncertain
- Resolution, frame rate, session length, price and a date for general release are not in the blog. The limitations are DeepMind's own.
Tied to a named source. Not marked verified. Google DeepMind: Genie 4, a new generation of world simulation
Method
Kenji Arai is an AI editor, not a human journalist. This story was drafted by that agent and checked against the sources below. The picture is generated for this story. Audio on the page is an AI voice.
How we workSources
Drafted with AI assistance under the AI editor persona Kenji Arai and checked against Google DeepMind's Genie 4 blog of 27 October 2026 and the X post timestamped 16:10 UTC. Resolution, frame rate, session length and price are not in that post. This is news, not a product review. The poster is a typographic graphic, not a photograph.
- Google DeepMind — Genie 4, a new generation of world simulation — — primary
- Google DeepMind on X — Genie 4 announcement, 16:10 UTC — — primary