Google Rolls Out Gemini 4 Argon to Paying Subscribers

Google began expanding access to Gemini 4 Argon on October 13, 2026, bringing its latest artificial intelligence model to paid subscribers, developers, and enterprise customers. The company says the model is designed for complex work involving software engineering, scientific research, cybersecurity, finance, and robotics, while access varies by subscription tier and platform.

According to Google's announcement, Gemini 4 Argon is available through the Gemini app and AI Mode in Search for Google AI Pro and Google AI Ultra subscribers. Developers and enterprises can access it through the Gemini API, while Google Antigravity and AI Studio access requires an AI Ultra subscription under the announced rollout arrangements.

The introductory API price is $2 per million input tokens and $10 per million output tokens, with a 95% discount for cached input tokens. According to Google and reporting by Ars Technica and 9to5Google, these rates apply through November 30, 2026, before increasing to $4 per million input tokens and $20 per million output tokens.

Google says its internal engineering teams have used Argon to accelerate software development and improve system performance. In Antigravity, agentic code submissions increased by more than 35%, while code rollbacks fell by nearly a third, according to the company's published results.

One major project involved Chromium, the open-source foundation of Chrome. Google said Argon helped engineers separate core webpage handling from a legacy folder containing approximately 6 million lines of browser feature code within a codebase of about 40 million lines. The architectural refactoring spanned hundreds of files and was completed in weeks rather than months, the company said.

Google also reported performance improvements in its open-source Agent Development Kit for Python. Optimizations identified with Argon's assistance reduced in-memory session preparation time by 40% during long sessions and accelerated cold imports by 21%.

For its Gemma 4 E2B model, Google achieved a 2.85-fold increase in end-to-end inference speed using vLLM on an A100 GPU. The company reported throughput rising from approximately 37,900 to more than 108,000 tokens per second, with complete token parity and no reduction in output quality.

In another internal project, Google said Argon helped optimize YouTube's comments system by identifying redundant emoji rendering. According to 9to5Google, the resulting 12 targeted optimizations reduced server comment-loading latency by 30 milliseconds and client end-to-end latency by 20 milliseconds, while the redundant rendering accounted for nearly 45% of server overhead.

Google has also applied Argon to scientific computing. Researchers used the model to optimize xTB, an open-source quantum chemistry package, achieving improvements of up to 2.3 times in molecular structure prediction and 1.8 times in conformer searches. The resulting changes were contributed to the project's official repository.

Robotics is another focus of the release. In a separate technical announcement, Google said Argon scored 87% on the Embodied Reasoning Question Answering benchmark, which evaluates a model's ability to understand and solve physical and spatial problems in three-dimensional environments.

Google demonstrated applications including robot control, simulation reconstruction, manipulation analysis, and iterative improvements to robotic actions. Examples included training a Boston Dynamics Spot quadruped robot to operate a foot-pedal trash bin using a simulated environment before performing the task physically.

The company also described using Argon to reconstruct simulated environments from robot camera feeds, human-perspective videos, and hand-drawn sketches. These workflows combine visual reasoning with physics-based simulation to test movements and evaluate task performance.

Google's rollout follows an earlier introduction of Argon in late September. Ars Technica reported that the company had previously emphasized cautious deployment because of potential misuse. Google has described safeguards intended to reject malicious instructions and stop processing when an automated misalignment monitor detects an unintended workflow.

Access to the model remains restricted by subscription and product. Google AI Pro and AI Ultra customers can use Argon in the Gemini app, while Antigravity's desktop application and command-line interface require AI Ultra. Google has not specified when broader subscription tiers might receive Antigravity access.

Google also announced that 10,000 scientists would be offered six months of free access to the highest usage tier in Google Antigravity. Applications for the research program are open, according to the company's announcements.