Genie 4
modelYour notes
Google DeepMind's fourth-generation world model, announced October 27, 2026, which generates interactive environments in real time that a user can explore and steer with natural language, changing the weather or adding objects mid-session. The headline change is that it is natively multimodal: music, ambient sound and adaptive character dialogue are generated together with the world. Google also cites more realistic physics, such as aerodynamics and collisions, and more natural human characters. It is built on the Gemini family: Gemini Omni anchors visual appearance and behaviour, other Gemini models keep the world coherent, and Gemini 3.8 Flash can act as a guiding agent that plans steps and checks progress inside the world.
It launched as a research preview for trusted testers, with access "coming soon" for Google AI Ultra subscribers in Project Genie and for early-access Gemini API and AI Studio partners. Output carries SynthID watermarks. Google lists the open problems itself: visual detail and object permanence drift over long sessions or after repeated edits, and real-time audio accuracy, legible text, spatial accuracy and long-term consistency remain research areas. The announcement gives no resolution, session length, parameter count, architecture or paper, and no quantitative comparison with Genie 3.