Google DeepMind Launches Gemini 4 Argon, Limited to Fairwind Program

Gemini 4 Argon arrives after Google DeepMind's last larger-than-Flash model in February, 3.1 Pro, and after successive incremental 3.x Flash versions and a big GDM management shakeup last month. The launch addresses the question of when GDM would catch up to peers who had launched Fable and Astra class models in the meantime. Argon's benchmarks are described as very respectable, with SOTA in 13 of 19 credible benchmarks, but access is currently limited to a cybersecurity preview for government users and trusted cyber defenders.

Beyond benchmarks, Google reports internal deployments where Argon agents freed more than 300 TiB of data-center memory and are migrating more than 800K lines of C/C++ kernel code to Rust. Agents also replaced 32K lines of SIMD code with safe Rust, making the existing Rust port 2.7x faster with identical output. These internal results suggest the model may be useful for large-scale code modernization, though the body does not specify how these were verified.

The outcome hinges on whether Google can refine guardrails and open access as soon as possible, and whether the introductory pricing discount, which brings the cost to $2/$10 per 1M input/output tokens with no end date announced, continues. Skepticism from some observers about the published numbers and a reported 19.6% on Harvey's legal benchmark trailing Muse Spark 1.2's listed 25.42% suggest the benchmarks may face scrutiny. For enterprise security teams and government cyber defenders in the Fairwind Program, early access could provide a head start on evaluating the model's capabilities.