Not established from the available sources.
GOOGLE DEEPMIND · MODEL RELEASE TRACKER
Genie 4
Genie 4 is Google DeepMind’s new real-time interactive world model that generates photorealistic scenes with native audio and can be steered via natural-language prompts; it is launching as a limited research preview for trusted testers.CURRENT SNAPSHOT3/5 DIMENSIONS WITH DATA
The dimensions that change the decision.
Typical scene length / consistencyGenie 4 can maintain a consistent scene for up to about two minutes — roughly double the length reported for Genie 3 in DeepMind/coverage — before details and object permanence begin to degrade.
Modalities (multimodal, native audio, dialogue)Genie 4 is natively multimodal: it generates photorealistic visuals together with native audio — including music, ambient sounds, and interactive character dialogue — produced directly by the model.
Not established from the available sources.
Rollout and initial accessDeepMind has started rolling out Genie 4 to a limited group of trusted testers; it will be made available to Google AI Ultra subscribers in Project Genie and to early access partners of the Gemini API and Google AI Studio (waitlist open).
VERIFIABLE FACTS
Every value stays attached to a source and date.
RELEASE · Announcement / release statusDEVELOPER CLAIM
Genie 4 was announced as a new world model and is being released as a research preview.
AVAILABILITY · Rollout and initial accessDEVELOPER CLAIM
DeepMind has started rolling out Genie 4 to a limited group of trusted testers; it will be made available to Google AI Ultra subscribers in Project Genie and to early access partners of the Gemini API and Google AI Studio (waitlist open).
CAPABILITIES · Real-time interactive simulation and steeringDEVELOPER CLAIM
Genie 4 simulates environments frame-by-frame in real time and lets users steer scenes mid-simulation using natural-language prompts (e.g., change weather, move camera, add people/objects); it can set events in motion and simulate subsequent world responses.
MODALITIES · Modalities (multimodal, native audio, dialogue)DEVELOPER CLAIM
Genie 4 is natively multimodal: it generates photorealistic visuals together with native audio — including music, ambient sounds, and interactive character dialogue — produced directly by the model.
CAPABILITIES · Photorealism, movement, and physics improvementsDEVELOPER CLAIM
Genie 4 delivers enhanced photorealism, more fluid character movement, and improved physics compared with its predecessor (Genie 3); DeepMind highlights more realistic people and scene behavior.
CAPABILITIES · Integration with Gemini family for planning and steeringDEVELOPER CLAIM
Genie 4 builds on capabilities from the Gemini family: Gemini Omni anchors simulations in real-world appearance and behavior, while other Gemini models (e.g., Gemini 3.8 Flash, Gemini 4 Argon in tests) plan and steer simulations to maintain coherence.
CONTEXT WINDOW · Typical scene length / consistencyDEVELOPER CLAIM
Genie 4 can maintain a consistent scene for up to about two minutes — roughly double the length reported for Genie 3 in DeepMind/coverage — before details and object permanence begin to degrade.
LIMITATIONS · Known limitationsDEVELOPER CLAIM
DeepMind describes Genie 4 as an experimental research model with limitations in longer sessions and complex physics; as simulations run longer, details and object permanence can shift and complex physics remain challenging.
WHAT CHANGED