Google has taken a significant step towards the long-term goal of artificial general intelligence (AGI) with the unveiling of Genie 3, a cutting-edge AI system capable of generating realistic virtual environments from simple text prompts. According to Google DeepMind, the company’s artificial intelligence research division, the model is a foundational piece of technology designed to help train robots, autonomous vehicles, and intelligent agents by simulating the real world.
Described as a world model, Genie 3 can create immersive, physics-based scenarios such as skiing down a mountain or navigating a warehouse, all generated instantly in response to user instructions. The model allows for rapid modifications to environments; for example, users could add a herd of deer to a ski slope simply by typing a new prompt.
Although Genie 3 has not yet been released to the public, journalists were shown examples of its capabilities earlier this week. The quality of its simulations was compared favourably to Google’s most advanced video model, Veo 3, although Genie’s environments run for minutes rather than Veo’s current limit of eight seconds.
DeepMind said Genie 3 is not just about visual spectacle; it is a tool that could help AI agents learn by interacting with realistic digital worlds. Such capability is viewed as a critical milestone on the road to AGI, a hypothetical form of AI that can perform a wide range of cognitive tasks on par with humans, unlike today’s more narrowly focused systems.
“We expect this technology to play a critical role as we push toward AGI, and agents play a greater role in the world,” DeepMind said in a statement.
The model could also aid in human learning and exploration, Google noted, allowing people to experience virtual environments for training purposes or even recreational simulations like skiing or hiking around a mountain lake.
Despite its potential, Genie 3 remains in development, with Google citing limitations that prevent full public deployment for now. No release date has been given.
The announcement comes as competition in the AI sector continues to intensify. Over the weekend, OpenAI’s CEO, Sam Altman, shared a glimpse of what appeared to be the company’s latest model, GPT-5, suggesting the race toward next-generation AI is far from slowing down.
World models such as Genie 3 are considered essential not only for AGI but also for the development of physical AI systems. Google envisions a future where realistic warehouse simulations, for instance, can train robotic systems in how to navigate and interact with humans and objects in complex environments. These systems could then transfer their learned skills from the virtual world to real-world scenarios.
Andrew Rogoyski, of the Institute for People-Centred AI at the University of Surrey, said: “If you give a disembodied AI the ability to be embodied, albeit virtually, then the AI can explore the world, or a world and grow in capabilities as a result.”