Master Google DeepMind's Genie 3 world model: Generate interactive 3D environments in real-time, with promptable world events and consistent memory for several minutes.
Genie 3 is Google DeepMind's latest general-purpose world model that generates navigable, interactive 3D environments in real time at 24 fps / 720p, maintaining visual and spatial consistency for several minutes. Released in 2025 as a stepping stone toward AGI.
⚠️ Current Status: Limited research preview for selected academics and creators. Not publicly available yet. DeepMind is working to expand testing access.
If you have been granted access to the research preview:
Practical patterns for testing Genie 3's capabilities. Click each pattern to see full details.
Currently limited to research preview for selected partners. Follow DeepMind's official channels for updates on expanded access.
A Google DeepMind world model that generates interactive 3D worlds in real time (720p/24fps) with promptable world events and minutes-long consistency.
Yes, Genie 3 maintains scene consistency for several minutes and has approximately 1-minute memory window for object persistence when out of view.
Yes, DeepMind has tested Genie 3 with their SIMA agent for goal-oriented tasks in generated worlds, positioning it as crucial for AGI development.
No public release date announced. DeepMind states they're working to expand testing to additional users in the future.