Google DeepMind is working to transform its most advanced multimodal foundation model, Gemini 2.5 Pro, into what it calls a 'world model' - an AI system capable of understanding and simulating aspects of the real world in ways that mimic human cognitive processes.
Announced at Google I/O 2025, this ambitious initiative aims to create an AI that can make plans and imagine new experiences by modeling how the world works. According to DeepMind CEO Demis Hassabis, this capability represents a fundamental advancement in artificial intelligence that goes beyond simple classification and prediction.
"This is why we're working to extend our best multimodal foundation model, Gemini 2.5 Pro, to become a 'world model' that can make plans and imagine new experiences by understanding and simulating aspects of the world, just as the brain does," Hassabis explained in a recent blog post.
Google sees early evidence of these world-modeling capabilities already emerging in Gemini's ability to represent and simulate natural environments, understand intuitive physics through its Veo model, and teach robots to grasp objects and follow instructions through Gemini Robotics.
The company describes this development as a critical step toward creating a "universal AI assistant" - an intelligent system that understands context and can plan and take action across any device. Google's ultimate vision is to transform the Gemini app into an assistant that can perform everyday tasks, handle administrative work, and provide personalized recommendations.
This world model capability builds upon Gemini 2.5 Pro's already impressive reasoning abilities. The model currently leads on benchmarks for complex math, science, and coding tasks, and ranks first on the LMArena and WebDev Arena leaderboards, which measure human preferences for AI interactions.
Google plans to integrate these world model capabilities with technologies from Project Mariner and Project Astra, which focus on multitasking and visual understanding respectively. The enhanced Gemini 2.5 Pro is expected to be generally available in June 2025, following its current preview phase.