ModelGemini

Google launches Gemini 4 Argon for robotics

Original title: Gemini 4 Argon for Robotics

Plain language

What it is: Google introduced Gemini 4 Argon, an AI model designed for robotics. It improves physical understanding, code generation, and visual reasoning.
How it works: The model plans actions for robots, creates digital simulation twins from videos or sketches, and controls robots in real time. It processes visual data and makes step-by-step decisions to solve physical tasks.
Where it is available: The model applies to robotics systems and simulation environments. Specific details about commercial availability, APIs, or regions are not stated in the source.
Example: Controlling a four-legged robot to pick up soda cans and drop them into a foot-pedal trash bin after practicing the movement in a digital simulation.

What it is: Gemini 4 Argon was announced, a model for robotics that scores 87% on the ERQA benchmark for spatial and physical reasoning in 3D environments. It enhances physical understanding, code-as-policies generation, and visual-semantic reasoning.
How it works: It operates using zero-shot control, Real2Sim reconstruction, and closed-loop steering of video world models. It translates visual coordinates and sensor data into motion primitives and simulation assets.
Where it is available: The model is applied within the robotics stack, MuJoCo simulations, and the Gemini Omni world model. Public API availability, pricing plans, or regions are not stated in the source.
Example: Reconstructing a calibrated 3D MuJoCo simulation from a real robot teleoperation video, matching the simulated contact forces with the logged tactile sensor data.

The headline and explanation on this page are based on Google's post. The full details are in the original.

Gemini 4 Argon for Robotics