Gemini 4 Argon

Google has unveiled Gemini 4 Argon, a new frontier AI model designed to assist cyber defenders in their efforts against increasingly sophisticated cyberattacks. This powerful model, which scales up the company's previously released Gemini 4, delivers frontier performance in complex workflows across software engineering, enterprise knowledge work, and cybersecurity defense.

Gemini 4 Argon is being rolled out to a select group of trusted cyber defenders through Google's Fairwind Program. The model's expanded output token limit to 1 million tokens, up from 64K, allows it to sustain deep reasoning across longer, multi-step tasks. This enables it to excel in a range of enterprise workflows, from coding to legal and financial tasks.

In tests, Argon demonstrated leading performance on various benchmarks, including DeepSWE v1.1, which measures software engineering tasks, and the Vals Index, which evaluates the economic impact of models in finance, coding, legal, and tax work. It also outperformed other models in specialized cybersecurity tasks, such as finding and remediating security vulnerabilities.

Argon's unique strength lies in its ability to autonomously find, validate, and patch critical software vulnerabilities, making it an invaluable tool for cyber defenders. A demonstration of this capability involved Argon uncovering a critical vulnerability in healthcare software used by hospitals worldwide, a risk that previous frontier models had missed.

To ensure the safe use of this powerful model, Google has implemented robust safeguards to prevent misuse, protect against prompt injection attacks, and monitor for misalignment. These measures include refusing harmful requests, improving robustness against misuse, and deploying mitigations to stop execution when Argon's actions diverge from user intentions.