Google Unveils Gemini 4 Argon AI Model for Cybersecurity Defenders

Google has rolled out its new Gemini 4 Argon AI model to select cybersecurity defenders as part of its Fairwind Program. The model, available only to vetted members of the program, is designed to surpass rival models in various benchmarks.

Gemini 4 Argon boasts improved performance in long-horizon software engineering tasks, with an output limit of 1 million tokens, and demonstrates strong results on DeepSWE and AutomationBench. It also exhibits enhanced resistance to indirect prompt injection and features separate monitors to prevent unintended actions.

Google employees are already utilizing the model for large-scale engineering projects, such as code migration and optimizing existing codebases, resulting in significant performance improvements and memory savings.