Google’s DeepMind has just unveiled Genie 3, a next‑generation world model capable of creating immersive, interactive 3D environments in real time.
This foundation model marks a breakthrough in AI, simulating dynamic worlds at 720p resolution and 24 frames per second, with visual memory that preserves consistency for several minutes.
Unlike its predecessor Genie 2, which only supported brief 10 to 20 seconds of interaction, Genie 3 enables continuous, coherent navigation over extended time spans.
DeepMind describes Genie 3 as the first truly real‑time interactive general‑purpose world model, able to generate both lifelike and fantastical worlds from simple text prompts located anywhere between realistic and imaginary settings. The model also supports prompt‑driven events—users can change weather, introduce objects or alter scenes on the fly.
DeepMind is positioning Genie 3 as a crucial technology for advancing toward artificial general intelligence (AGI). According to the company, the ability to simulate environments where embodied agents can navigate and reason is key to training systems that behave intelligently across diverse tasks.
Potential applications include training robots in virtual warehouses, prototyping complex educational or entertainment scenarios, and enabling designers to visualize rich environments instantly. However, DeepMind notes several limitations i.e. the agent interaction range remains narrow, multi-agent dynamics are still hard to model, and continuous simulation remains limited to minutes rather than hours.
Genie 3 is currently offered as a controlled research preview to select academics and creators. DeepMind says this limited rollout allows them to evaluate risks and gather feedback before wider availability.