Gemini 4 Argon for Robotics
deepmind ·
DeepMind has introduced Gemini 4 Argon, a new model designed to significantly advance physical understanding, code generation, and visual-semantic reasoning. This release aims to unlock new workflows throughout the robotics stack and accelerate the development of physical artificial general intelligence. The model's capabilities are demonstrated through its performance on the Embodied Reasoning (ERQA) benchmark, which evaluates how well models solve physical and spatial problems in 3D environments. By relying on pure reasoning, Gemini 4 Argon achieved a score of 87% on this benchmark. This result positions it as the top-performing model, surpassing all other frontier models currently available. The announcement highlights a major step change in the model's ability to interact with and understand the physical world. These improvements are intended to support more complex robotic applications and tasks.
deepmind