Google DeepMind Releases Gemini 4 Argon for Robotics Applications

Disclosure: this post contains affiliate links. If you buy through them we may earn a commission at no extra cost to you. Learn more.

TL;DR: Google DeepMind released Gemini 4 Argon on October 13, 2026, a general-purpose physical reasoning model that scores 87% on the Embodied Reasoning (ERQA) benchmark. The model enables applications across the entire robotics lifecycle including robot control, Real2Sim reconstruction, physical understanding, and research self-improvement.

What's New

Google DeepMind has announced Gemini 4 Argon for Robotics, marking what the company describes as "a step change in physical understanding, code generation, and visual-semantic reasoning." Released on October 13, 2026, the model achieves 87% on the Embodied Reasoning (ERQA) benchmark, which tests a model's ability to understand and solve physical, spatial problems in 3D environments. According to DeepMind, Argon outperforms every other frontier model using pure reasoning.

As a general-purpose physical reasoning model, Argon can be applied across four key areas of the robotics lifecycle. For robot control, the model demonstrates mobile manipulation with MuJoCo practice, zero-shot robot control, and closed-loop steering of video world models. In one demonstration, Argon controls a Boston Dynamics Spot quadruped (12-DoF legs + 6-DoF arm) to learn a new skill: stepping on a foot-pedal trash bin to throw away soda cans. The model creates an articulated MuJoCo digital twin of the trash bin to practice moves before executing on physical hardware.

For zero-shot control, Argon can control a simple robotic arm to place a fallen mug upright and put a crumpled napkin inside, reasoning with a visual coordinate harness and action primitives over several turns. The model also demonstrates the ability to steer robots in the Gemini Omni World Model, rearranging scattered Scrabble tiles into a clean row by writing timestamped action prompts and inspecting intermediate rollout keyframes in a closed visual loop.

Real2Sim and Physical Understanding

Argon's Real2Sim capabilities enable reconstructing physically calibrated simulation twins from multiple sources including robot camera feeds, egocentric human video, or hand-drawn paper sketches. In one demonstration, Argon views a real robot teleoperation episode and develops a digital twin, obtaining or creating simulation assets that look and behave like real objects upon contact with dexterous hands. The model tunes assets and scenes to match simulation visuals and contact events with real recorded data in an iterative visual-reasoning loop.

For egocentric human video, Argon can identify task-relevant objects, replace crude bounding boxes with compound convex-decomposed meshes for concave containers, and retarget human hand trajectories onto a 76-DoF bimanual humanoid with 22-DoF five-fingered hands in MuJoCo contact physics at 500 Hz.

For physical understanding, Argon performs agentic multimodal hindsight labeling including detailed understanding of dexterity, dense value functions, and progress estimation. The model also supports research and self-improvement through action improvement via simulation and iterative self-improvement on hardware.

Pricing and Availability

Pricing and availability information for Gemini 4 Argon was not disclosed in the announcement.

How to Try It

The announcement includes detailed prompt examples across various robotics applications, but specific instructions for accessing or trying Gemini 4 Argon were not provided.

Who It's For

Gemini 4 Argon is designed for robotics researchers, developers, and engineers working across the entire robotics stack. The model's capabilities span from low-level robot control to high-level task planning, making it relevant for teams building mobile manipulation systems, developing simulation environments, or working on physical AI applications.

Alternatives

While specific alternatives were not mentioned in the announcement, the release positions Argon as outperforming "every other frontier model" on the Embodied Reasoning benchmark. Organizations working on robotics workflows may also consider workflow automation platforms like Make for integrating AI capabilities into their robotics pipelines.