Gemini 4 Argon scores 87% on Embodied Reasoning, outperforming every frontier model
Gemini 4 Argon for Robotics
Google's Gemini 4 Argon brings a step change in physical understanding, code generation, and visual-semantic reasoning for robotics. It scores 87% on the Embodied Reasoning (ERQA) benchmark, beating all other frontier models. The model spans the robotics lifecycle: controlling a Boston Dynamics Spot to dispose of cans, reconstructing MuJoCo digital twins from robot or human video, steering video world models, and labeling manipulation data. It also turns hand-drawn sketches into verified humanoid simulations.
Using pure reasoning, Argon scores 87%, outperforming every other frontier model.