Google makes Gemini 4 Argon available to paid users, showcasing robot simulation and training capabilities

Article

Google Gemini 4 Argon officially launches: watching a video teaches robots to throw beanbags, hitting the target consecutively 7 times

Beating AI Breaking News: Google has officially made Gemini 4 Argon available to paid users worldwide. Previously, the model was provided only to select cybersecurity partners. The most eye-catching demonstration shows a robot watching a video of a person throwing beanbags, identifying the positions on the field, calculating full-body joint movements, and writing a physics simulation program on its own. After simulation and sim-to-real transfer, Apollo, a humanoid robot from U.S. robotics company Apptronik, hit the target consecutively in a real-world setting 7 times.

Google says this process reduces the manual simulation modeling that originally took weeks to just a few hours. In other demonstrations, Argon built a simulated environment for Boston Dynamics' Spot robot dog, allowing it to practice stepping on trash-can lids to open them and putting cans inside. It can also generate a simulated scene from a photo of a tabletop and synthesize more than 4500 training-data segments. After using these data to fine-tune a robot action model, the real-robot task success rate increased from 56%to 75%; the time required fell from 35 seconds to 24 seconds. In the ERQA evaluation, which tests three-dimensional spatial and physical understanding, Argon scored 87%, outperforming other frontier models.

Currently, Google AI Pro and Ultra users can use Argon in the Gemini app, but it is available only to Ultra users in Antigravity. Developers can also access it through a paid API API.API launch pricing is Token 2 USD per million input 10 USD, 5.5、GPT-6.1 per million output 5.1, matching Claude Sonnet GPT-6 Sol and only one-fifth of Claude Fable November 30 and 5.5 Astra. However, this discount lasts only until https://m.theblockbeats.info/flash/371663

Key points

The report says Google has officially made Gemini 4 Argon available to paid users worldwide; it was previously provided only to select cybersecurity partners.
After watching a video of a person throwing beanbags, Argon can identify the positions on the field, calculate full-body joint movements, and write a physics simulation program on its own.
After simulation and sim-to-real transfer, Apptronik's Apollo humanoid robot hit the target consecutively in a real-world setting 7 times.
The report says this process reduces the time required for manual simulation modeling from weeks to a few hours.
Argon can build a simulated environment for the Spot robot dog, generate a simulated scene from a tabletop photo, and synthesize more than 4500 training-data segments.
After fine-tuning the robot action model with the relevant training data, the real-robot task success rate increased from 56%to 75%, while the time required fell from 35 seconds to 24 seconds.
Argon scored ERQA in the 87%。
Google AI Pro and Ultra users can use Argon in the Gemini app; Antigravity is available only to Ultra users, and developers can access it through a paid API API.
API The launch price is Token per million 2 USD for input and 10 USD for output, with the discount lasting until November 30, after which the price doubles.

AI-assisted interpretation

The following is analysis, separate from reported facts. Verify important claims independently.

This report describes a model reportedly called Gemini 4 Argon. It is used to convert videos or photos into robot simulation environments, training data, and action plans, which are then transferred to real robots for execution.

Why it matters to readers

If the report is accurate, Argon could make robot training and simulation preparation faster and, through the Gemini app and paid API expand the scope of access.

It can be understood as follows: the model first understands the task from a video or photo, then trains the robot in a virtual environment, and finally has the real robot attempt to perform it.

Risks and unknowns

The evidence does not provide an official Google announcement or an original link to the robot demonstration.
The model capabilities, evaluation scores, success rates, pricing, and availability described in the report still require further verification.
The evidence does not specify the actual regions where it is available,API limitations, or whether the robot demonstration can be reproduced by third parties.

Google makes Gemini 4 Argon available to paid users, showcasing robot simulation and training capabilities

Article

AI-assisted interpretation

Why it matters to readers

Risks and unknowns

Related Developments

Related concepts

Keep following

Security Center