Google Gates Access to New Gemini 4 Argon Model

Google has unveiled Gemini 4 Argon, a new frontier AI model designed for high-performance complex workflows in cybersecurity, software engineering, and enterprise knowledge work such as finance and legal, according to The Verge.

Despite the capabilities, Google is restricting initial access. Koray Kavukcuoglu, Google DeepMind SVP and chief AI architect, told The Verge that the company is limiting the model to a "set of trusted cyber defenders" and is participating in a voluntary U.S. government process for pre-release access. Ars Technica reports that partners in Google's Fairwind Program can access the model's cyberdefense tools, and the security firm Wiz has already used Argon to identify a critical vulnerability in systems used by global hospitals.

While the public waits, Google is already utilizing the model internally. Ars Technica reports that Argon helped Google save 300 TiB of memory across data centers and migrated over 800,000 lines of code in the Fuchsia OS Zircon kernel to Rust.

Google cites the need to monitor for misalignment and defend against prompt injection attacks as reasons for the phased rollout. The company has not provided a specific timeline for general availability, though Ars Technica notes that access will eventually extend to Google AI Ultra subscribers and paid API users. Google has set initial API rates at $10 per million output tokens and $2 per million input tokens.