Google unveils Gemini 4 Argon: limited access for now

Google recently presented its latest AI model, Gemini 4 Argon. This model is designed to compete with the most powerful systems from competitors like OpenAI and Anthropic. It specializes in complex tasks ranging from software development to cybersecurity. Despite Argon's strong performance in these areas, Google has decided to limit access to the model for a wider audience for the time being.

The unveiling of Gemini 4 Argon took place on September 30 and was announced by CEO Sundar Pichai. He emphasized that, given the intense discussions about the capabilities of the new model, it was important to provide an early glimpse of Argon. Pichai stated that the model offers significant advantages in executing complex workflows and excels in digital defense and technical development. The model is already being extensively used within Google itself.

The limited access to Gemini 4 Argon primarily targets selected cybersecurity partners participating in the Fairwind Program. These trusted organizations receive the tools to identify and address security vulnerabilities. Google is taking a cautious approach, as a model that detects vulnerabilities can potentially also be misused to attack systems. Before the model is made widely available, Google wants to gather feedback and further test the security measures. Additionally, the U.S. government is prioritized for access through an evaluation process.

Internally, Google is already using Argon to optimize quantum calculations and convert large amounts of C/C++ code to Rust. The model has also led to significant efficiency gains in their data centers. The introductory price for the model is set at $2 per million input tokens and $10 per million output tokens, with additional benefits for cached input. While Google claims strong benchmark results, it remains unclear when the broader release will take place, leaving users to assess their own experiences with the model later.

Read the full article from AI Insider.