Google Unveils Gemini 4 Argon for Robotics

Google has released Gemini 4 Argon, a groundbreaking model designed for robotics, enhancing physical understanding, code generation, and visual-semantic reasoning. This general-purpose physical reasoning model excels in the Embodied Reasoning (ERQA) benchmark, achieving an impressive 87% score. Argon is applicable across the robotics lifecycle, enabling workflows like mobile manipulation, real-to-sim reconstruction, and agentic multimodal hindsight labeling. Developers can leverage Argon through various applications, including robot control with MuJoCo, zero-shot robot control, and closed-loop steering of video world models. According to Google, Argon outperforms other frontier models in solving physical and spatial problems within 3D environments. The model is available for use in diverse scenarios, promising to accelerate progress toward physical AGI in robotics.