Google has unveiled Gemini 4 Argon, its latest frontier AI model, effectively skipping the anticipated Gemini 3.5 Pro. While the company claims industry-leading performance in coding, knowledge work, and cybersecurity, the model remains restricted to a small group of testers for now.
Internal Efficiency and Code Migration
Google reports that its own engineers are already using Argon extensively. The model has reportedly utilized fleet-wide telemetry data to save 300 TiB of memory across Google’s data centers. Furthermore, Argon-powered agents have been tasked with migrating C/C++ codebases to Rust, including more than 800,000 lines in the Fuchsia OS Zircon kernel and core libraries like re2 and libgav1.
Benchmarks and Technical Specifications
In software engineering tests, Gemini 4 Argon reached 77.9 percent on the DeepSWE v1.1 benchmark, outperforming GPT-6 Astra, Fable 5.1, and Opus 5.5. A significant technical leap is the model's output limit, which has been raised to 1 million tokens, up from 64,000 in previous versions. Google has also announced API pricing at $2 per million input tokens and $10 per million output tokens, with a 95% discount for cached inputs.
Cyberdefense and Safety Protocols
The primary focus for Gemini 4 is currently cyberdefense. Through the Fairwind Program, partners like Wiz have already used the model to identify a critical vulnerability in hospital systems that could expose personal data. To address safety concerns, Google has implemented systems to monitor the model’s "chain-of-thought," allowing it to intervene if the AI attempts to step out of predefined safety bounds.
