Google previews Genie 4 world model

Genie 4 is the latest in a series of world models Google launched in 2024, and the company says it improves on its predecessor in rendering fluid movements and other complex details. The model relies on other Google systems under the hood: Gemini Omni, a video generator introduced in May, helps keep visual assets realistic, while Gemini 3.8 Flash, a cost-efficient reasoning model, makes individual frames consistent. According to the DeepMind staffers, Gemini observes the scene, breaks goals into steps and adapts its plan as the world shifts or the user intervenes.

Beyond consumer and developer access, DeepMind is studying robotics applications that combine Genie 4 with Gemini 4 Argon, Google's flagship reasoning-optimized large language model. In that setup, Argon can turn a brief request into dozens of detailed Genie 4 prompts describing each step of a workflow, and more precise prompts appear to improve rendering quality—which the company suggests makes the simulations more useful for training AI to control robots. Google is testing Genie 4 with a limited number of early adopters before wider availability through Google AI Ultra, the Gemini API and Google AI Studio.