Hot AI News

Gemini 4 Hydrogen can set up a robot arm from a USB cable in about 10 minutes

Google says Gemini 4 Hydrogen matches Argon on a real-arm robotics benchmark at a lower cost per token, and a new Antigravity plugin covers setup, simulation and control.

Gemini 4 Hydrogen can set up a robot arm from a USB cable in about 10 minutes
Image: Google

Google has released a Robotics plugin for its Antigravity workbench, built around Gemini 4 Hydrogen, its faster model. According to the Google Developers Blog, Hydrogen scores 58 on the Artificial Analysis Robotics 0.1 benchmark, similar to Argon, at a lower cost per token.

What the plugin and the benchmark cover

Google describes Hydrogen as a model with strong spatial and embodied reasoning that is built for speed and efficiency. A developer states an objective in everyday language and supplies a camera view; the model then works out a plan, produces and executes the code that drives the arm, and keeps refining its approach until the task is finished. The company says this direct control of a robot arm carries over from Gemini 4 Argon and Carbon.

Benchmark chart from Google's Gemini 4 Hydrogen robotics announcement
Image: Google

The Robotics 0.1 benchmark from Artificial Analysis has the model control a real 6-axis arm through eight everyday tabletop tasks, such as pouring a drink or packing blocks into a box. Google reports the 58 score but does not publish Argon’s exact figure or the cost per token in the post, so the size of the saving cannot be checked from the source.

Google also says Hydrogen absorbs the skills of its Gemini Robotics ER 2 embodied reasoning models. On the public ERQA benchmark it reports that Hydrogen beats ER 2 on spatial tasks such as multi-view correspondence and 3D trajectory reasoning. Developers who use ER 2 today are advised to test Hydrogen for those workloads.

Gemini 4 Hydrogen can set up a robot arm from a USB cable in about 10 minutes
Image: Google

The plugin packages four groups of skills:

  • Robot bring-up: driver installation, camera discovery, arm calibration and safe-motion checks.
  • MuJoCo inside Antigravity: building and previewing simulated scenes and assets, including from links, documents or photos.
  • Perception and control: finding objects, planning grasps and closed-loop control.
  • Application building: generating the controller and simulation for a full real-to-sim-to-real project.
Advertisements

Demos range from a ten-minute setup to chess

Google’s examples are its own demonstrations, not independent tests. In the first, an I2RT YAM arm and a depth camera are plugged into a laptop, and the arm is calibrated in about 10 minutes. Google says similar flows brought up a low-cost SO-ARM101 and an industrial FR3 Duo.

Gemini 4 Hydrogen can set up a robot arm from a USB cable in about 10 minutes
Image: Google

A snack-sorting conveyor went from a whiteboard sketch to a working sorter in about an hour, with Hydrogen building the MuJoCo simulation, validating the grasping policy across runs and then deploying it to the real arm. A soap-shop storefront connected to two YAM Pro arms that pack orders and pull a printed note from a printer. An FR3 Duo simulated a chess board and played against itself or human opponents.

Google frames the pitch in terms of time: a developer with a newly unboxed robot could have a functioning application running within a single afternoon. That is the company’s claim, and the post gives no failure rates, no number of attempts and no hardware list beyond the arms shown.

Hydrogen’s robotics debut follows a fast rollout

The robotics push is the latest step in a quick spread of the model. Hydrogen arrived in GitHub Copilot as the successor to Gemini 3.8 Flash on November 14, and our weekly model roundup counted it among the releases of that week. A Google post on Bluesky on November 16 said Hydrogen is available to everyone across the Gemini app, Google Search and Google AI Studio.

On November 17 Sundar Pichai said Hydrogen powers Tasks in Gemini, which is expanding free to hundreds of millions of people in six countries, as we reported. Pichai’s recap of the previous week also listed Gemini Robotics 2.5 among Tuesday’s launches, so this is Google’s second robotics-related release within days. The same fast model now serves consumer chores, coding tools and physical control.

The pattern also explains the pricing emphasis. Google argues that Hydrogen’s efficiency makes it practical to run multi-step robot applications all day, which is where a cheaper token matters most. Whether the cost gap versus Argon is large enough to change developers’ choices depends on figures the post does not give.

Advertisements

Who can use it today, and what is missing

Developers select gemini-4-hydrogen in Antigravity, then open Customizations and search for Robotics on the Build With Google shelf. Google says Antigravity is for paid Google AI Pro or Ultra subscribers. The model is also in the Gemini API through the Interactions API, with day-one support in the google-genai SDKs, and in Google AI Studio for spatial prompts and live camera streams.

The post does not state regional limits, a price for the model, or which robots beyond the arms shown are supported. Safety is addressed only in passing: the bring-up skill verifies safe motion, but no details of how are given. Google invites feedback and shared projects on the plugin’s GitHub discussions page.

Sign Up for the Daily AI Pulse

One email a day. All the stories that matter.

Alex CarterAI News & Big Tech Correspondent

Alex Carter writes for Aiholics, keeping readers updated on the fast-paced world of AI and Big Tech.

Leave a Reply

Your email address will not be published. Required fields are marked *

You may also like