Google DeepMind released Genie 4, its real-time interactive world model, on October 27, 2026. It is DeepMind’s first real-time world model to generate audio natively, including music, ambient sound and dialogue, and users can change a running world with text, such as altering the weather, moving the camera or adding people and objects. It is available as a research preview to trusted testers.
Why it matters for robotics: DeepMind showed Genie 4 simulating robot tasks, with Gemini 4 Argon planning the steps. A world model that can stand in for the physical world is a source of training and test data.
Key Facts
- Released: Oct 27, 2026, research preview for trusted testers
- New: native audio; real-time text steering of the world
- Robotics demos: Apptronik Apollo playing checkers (with Gemini 4 Argon); UR5 arm dispensing liquid
- Limits: “short sessions”; detail drift on longer runs; SynthID on all output
What did the robot demos show?
In one, Gemini 4 Argon broke a task into steps that were fed to Genie 4 as text captions in real time, producing a simulated Apptronik Apollo humanoid playing checkers. In another, a UR5 arm dispensed liquid into a lab plate. DeepMind did not publish success metrics for either. Gemini 3.8 Flash can act as a guiding agent that watches a scene, breaks a goal into steps, checks each one and replans. We covered Argon’s role in Google’s robotics pitch for Gemini 4 Argon.
What are the limits?
DeepMind says Genie 4 keeps environments coherent over “short sessions” and gives no fixed duration; TechCrunch reported up to two minutes. Visual details and object permanence drift over longer runs and after repeated edits. DeepMind lists complex physical interactions, audio accuracy, legible text, spatial accuracy and long-term consistency as open problems. All output carries a SynthID watermark.
Who can use it?
Trusted testers now; DeepMind says it is coming to Google AI Ultra subscribers through Project Genie and to early-access partners in the Gemini API and AI Studio. Background: the world-models bet.
Frequently Asked
What is Genie 4?
Google DeepMind’s real-time interactive world model, released as a research preview on October 27, 2026. It generates video and audio and can be steered with text.
Can Genie 4 be used to train robots?
DeepMind showed it simulating robot tasks, including an Apptronik Apollo playing checkers and a UR5 arm dispensing liquid. It did not publish training results or success metrics.
How long can a Genie 4 session last?
DeepMind says it stays coherent over short sessions without giving a duration. TechCrunch reported up to two minutes.