Gemini 4 Argon: Google’s Powerful New AI Model for Coding and Cybersecurity

Google has introduced Gemini 4 Argon, its latest AI model aimed at demanding software engineering, enterprise automation and cybersecurity workloads. One of its biggest upgrades is the ability to generate up to 1 million output tokens, a major increase from the previous 64,000-token limit. Google says Argon recorded strong results across several benchmarks, including 77.9 percent on DeepSWE v1.1 for software engineering, 91.7 percent on LVBench for long-video understanding and 51.3 percent on AutomationBench for business tasks.

Google has also used Argon internally for engineering projects, including quantum computing and codebase migrations. In one project, the model helped create a Rust version of the libgav1 video decoder, which Google says runs 2.7 times faster than its existing Rust implementation while producing the same video output. For cybersecurity, Argon can identify, validate and patch software vulnerabilities and scored 68 percent on CWE-bench v1. Google is also testing safeguards against cyber misuse, CBRN-related risks, indirect prompt injection and unexpected model behaviour. The model is initially being tested through the Fairwind Program, with paid API customers and Google AI Ultra subscribers expected to receive access before developers, enterprises and consumers.