Google has announced Gemini 4 Argon, its newest frontier artificial intelligence model designed for complex professional workflows. The model is initially available to a select group of cyber defenders through Google's Fairwind Program, with wider release planned for developers, enterprises, and consumers.
What sets Argon apart
The most distinctive feature of Gemini 4 Argon is its industry-leading output token limit of 1 million tokens. When the model has the headroom to generate hundreds of thousands of tokens in a single trajectory, it adds a new level of depth in reasoning to solve tough problems, such as large-scale code migrations or multi-step financial research, in one go.
Google reports that Argon achieves state-of-the-art results on DeepSWE v1.1, a benchmark measuring performance on real-world software engineering tasks, with a score of 77.9 percent. The model also excels in legal and financial domains. For tasks requiring visual understanding, Argon scores 91.7 percent on LVBench, which tests long video comprehension.
Internally, Google has deployed Argon agents for several high-impact projects. These include optimizing quantum algorithms where Argon reduced spacetime resource requirements by 40 percent, analyzing fleet-wide telemetry to identify memory optimizations freeing over 300 terabytes of data center memory, and migrating hundreds of thousands of lines of C and C++ code to Rust.
Defensive cybersecurity focus
Argon is specifically trained for defensive cybersecurity operations. It can autonomously discover, validate, and patch software vulnerabilities. On Google's internal vulnerability benchmark spanning 20 programming languages, Argon demonstrated substantial improvements over previous models. On Wiz's black-box penetration testing benchmark, which tests analysis of live web systems without source code access, Argon outperformed earlier versions in attack surface discovery and proof-of-concept generation.
For trusted defenders, Google is releasing Argon without cybersecurity guardrails to allow full utilization of its defensive capabilities. Broader public release will follow additional safety testing and guardrail development.
Why this matters: The one million token output limit represents a significant expansion in AI capability, allowing models to generate extensive outputs for entire codebases, lengthy legal documents, or extended video content analysis. Combined with strong performance on professional benchmarks, this positions Argon as a tool for automating complex knowledge work that previously required human expertise spanning hours or days.