Google DeepMind releases Genie 3 world model

Google DeepMind opened public access to Genie 3, its world-generation model, through a consumer product called Project Genie — available to Google AI Ultra subscribers aged 18 and over in the United States, a subscription tier priced at $249.99. Genie 3 had first been shown in August 2025 as a limited research preview restricted to a small group of academics and creators; the January release was the model’s move from closed research access to a paid consumer product, fusing the world model with Google’s Nano Banana Pro image generator and Gemini.

The system generates an explorable, navigable 3D environment from a text or image prompt, rendering in real time at 720p and 24 frames per second and remaining broadly consistent for several minutes, with roughly a minute of visual memory. Project Genie sessions were capped at 60 seconds. DeepMind had acknowledged limitations in the original research preview, including a restricted action space, weak handling of multiple interacting agents, imprecise real-world geography and unreliable text rendering — constraints the January consumer release did not claim to have resolved.

Genie 3 belonged to a wider push by AI labs into “world models” — systems that simulate an interactive environment rather than generating a fixed video clip — positioned as a distinct research direction from both video generation and language modelling, with potential applications in game generation, robotics training and embodied-agent research through DeepMind’s SIMA project. Google said access would expand to more regions over time, without giving a timeline.