Gemini 4 Argon for Robotics

Google introduces Gemini 4 Argon, a model designed for robotics applications, showcasing advanced physical reasoning, code generation, and visual-semantic reasoning. It demonstrates capabilities in mobile manipulation, zero-shot control, and simulation-to-reality transfer, with performance on the Embodied Reasoning (ERQA) benchmark reaching 87%. The model supports tasks such as robot control, real-to-sim-to-real workflows, and iterative video world model steering.