Google Unveils Gemini 4 Argon With 1 Million-Token Output and Advanced Cybersecurity Capabilities

MOUNTAIN VIEW, CALIFORNIA, — October 1, 2026 — Google has unveiled Gemini 4 Argon, a new frontier artificial intelligence model designed for complex, long-running tasks in software engineering, enterprise work including legal and finance, and cybersecurity defense. The company announced the model on September 30 and said it is initially being provided to a limited group of trusted cyber defenders through its Fairwind program. (Google)

One of Argon’s headline features is an output limit of up to 1 million tokens, a substantial increase from the previous 64,000-token limit cited by Google. The company says the larger output capacity is intended to let the model sustain reasoning and execution across lengthy, multi-step tasks rather than being limited to shorter interactions. (Google)

Google says Argon has already been used internally for software engineering and other technical work. According to the company, Argon agents helped identify memory optimizations across Google’s data centers that have freed more than 300 tebibytes of memory after deployment, while other agents have been involved in large-scale C++ to Rust migrations. These figures are Google's own reported results and have not been independently audited. (Google)

Cybersecurity is a major part of the new model. Google says Argon can autonomously discover, validate and patch critical software vulnerabilities. In one early demonstration involving Wiz’s Scan for Good initiative, Google said Argon identified a serious vulnerability exposing sensitive personal information in healthcare software used by hospitals. The affected software was not publicly identified in Google's announcement. (Google)

The company is taking a phased approach to the release because of those capabilities. Google says trusted cyber defenders and its own internal teams will receive access first, while broader access to developers, enterprises and consumers will come later following additional testing and refinement of safety controls. (SecurityWeek)

Argon is also being introduced with an initial API price of $2 per million input tokens and $10 per million output tokens. Google says those introductory rates will later rise to $4 per million input tokens and $20 per million output tokens, while cached input tokens will receive a 95 percent discount from the standard input price. (Google)

Google's announcement places particular emphasis on controlling the risks of highly capable AI agents. The company says it is strengthening sandboxing, monitoring model behavior and testing defenses against indirect prompt injection before expanding access.

The launch marks another major step in the competition among leading AI developers over models capable of carrying out increasingly complex tasks with less human intervention. Google is positioning Argon around sustained reasoning, coding and cybersecurity rather than simply chatbot-style responses. (The Verge)

For now, Gemini 4 Argon remains a limited-release model. Google says wider availability will begin with paid API customers and Google AI Ultra subscribers before a broader public rollout. (Google)