Google DeepMind Reveals ‘Genie 3,’ a Groundbreaking AI World Model Advancing Towards Artificial General Intelligence
August 5, 2025 – Google DeepMind has announced the development of Genie 3, its latest foundational world model designed to simulate interactive three-dimensional environments in real-time, marking a significant leap towards artificial general intelligence (AGI), or human-like intelligence.
According to Shlomi Fruchter, a research director at DeepMind, Genie 3 represents the first real-time, interactive, general-purpose world model capable of generating extensive 3D environments from simple text prompts. Unlike previous models limited to narrow or domain-specific tasks, Genie 3 can create photorealistic and imaginary worlds at 720p resolution and 24 frames per second, sustained over several minutes—an improvement over its predecessor, Genie 2, which handled environments lasting only 10 to 20 seconds.
One of Genie 3’s key innovations is its ability to maintain physical consistency in its simulations over time because it ‘remembers’ previously generated environments. This memory component was not explicitly programmed but emerged from the model’s design, allowing it to simulate dynamic and evolving worlds with coherent physical rules and behaviors.
The model can also respond to “promptable world events,” giving users control to change the state of generated worlds through updates in the input prompts, thus enabling dynamic scenario creation which is crucial for training and testing AI agents in flexible environments.
Although still in research preview and not yet publicly accessible, Genie 3 builds on DeepMind’s prior innovations, including Genie 2 and the Veo 3 video generation model, which has an advanced understanding of physical dynamics. This progress aligns with Google’s broader AI initiatives, such as AlphaEarth, a model that aggregates trillions of satellite and radar observations to create detailed global maps, and the Gemini family of large language and multimodal models aimed at enhancing various AI capabilities.
DeepMind’s work with Genie 3 represents a crucial stepping stone in pursuing AGI—AI systems with broad, adaptable intelligence comparable to human understanding. By creating an AI system that can simulate detailed, persistent environments, researchers hope to enable the development of agents capable of learning, reasoning, and planning in ways not possible with current narrow AI models.
This announcement follows a series of Google AI updates in 2025, including the release of Gemini 2.5 models, Gemini CLI for developers, and new AI-powered tools across healthcare, education, and environmental monitoring, reinforcing Google’s commitment to advancing cutting-edge AI technologies responsibly.
Experts in the AI research community are watching closely as Genie 3 may prove to be a critical foundation for next-generation AI systems with applications ranging from robotics and gaming to scientific simulation and real-world problem solving.
Image Credit: Google DeepMind