Google DeepMind has introduced Gemini 4 Argon, a general-purpose physical reasoning model that the company says marks a step change in physical understanding, code generation, and visual-semantic reasoning. According to DeepMind, the model unlocks new workflows across the robotics stack and accelerates progress toward physical AGI.
On the Embodied Reasoning (ERQA) benchmark, which tests a model's ability to understand and solve physical, spatial problems in 3D environments, Argon scores 87% using pure reasoning, outperforming every other frontier model, DeepMind claims.
Argon can be applied across the entire robotics lifecycle. DeepMind shared representative prompts across four key areas: robot control, Real2Sim, physical understanding, and research and self-improvement. Examples include controlling a Boston Dynamics Spot quadruped to learn a new trash-disposal skill via a MuJoCo digital twin, steering robots in the Gemini Omni World Model to rearrange Scrabble tiles, and reconstructing physically calibrated simulation twins from robot camera feeds, egocentric human video, or hand-drawn sketches.
In physical understanding, Argon provides detailed segmentation and text labeling from video, identifies poor manipulation segments, and generates dense value function predictions. For research and self-improvement, Argon can generate new data in simulation to improve generalist vision-language-action models, boosting real-world physical robot success rate from 56% to 75% and reducing completion time from 35s to 24s in one example. It can also reproduce robotics papers in MuJoCo WebAssembly from a single prompt.
Gemini 4 Argon is available today across Google's developer platforms: Google AI Studio (for Google AI Ultra subscribers), the Gemini API (paid tier, using gemini-4-argon in the Interactions API with day-one support across google-genai SDKs), and Google Antigravity (for Google AI Ultra subscribers).