Most people cannot use Gemini 4 Argon yet. Per Google, its newest frontier model is going first to a small group of cyber defenders through the Fairwind Program, while developers, enterprises and consumers wait for a wider release. The staged approach is deliberate: Google is taking part in the U.S. government's voluntary process for pre-release model access and is still adjusting guardrails based on feedback from early testers. Paid API customers and Google AI Ultra subscribers are first in line once access opens, though no date has been given.
Google's performance claims are the usual mix of broad and narrow. It says the model ranks first on its Vals Index for finance, coding, legal and tax work, and scores 77.9% on DeepSWE v1.1, a test of long software engineering tasks. All of those results come from Google, and with access this limited few outsiders can check them. One change needs no benchmark to understand: the maximum output per response rises to 1 million tokens from 64,000, which gives the model room to reason through a problem in a single pass.
Google's own engineers supply the more concrete evidence. Argon agents combed through fleet-wide profiling data and applied memory optimizations that Google says will free more than 300 TiB once rolled out. In the libgav1 video decoder, agents replaced 32,000 lines of SIMD code in an existing Rust port with safe Rust that the compiler can vectorize. Google says the result runs 2.7 times faster than that port, with identical output. That one is the more interesting result of the two, because the code got safer and quicker at the same time. The agents are also working on C and C++ to Rust migrations, reaching more than 800,000 lines for the Fuchsia Zircon kernel, with automated and manual auditing before production, Google says.
The security work explains the caution. Google says it trained Gemini 4 Argon to find, validate and patch serious software vulnerabilities on its own, and that trusted defenders and its internal teams will get a version without cyber guardrails. Through its Scan for Good program, Wiz used the model to uncover a critical flaw exposing personal data in healthcare software used by hospitals, a risk Google says earlier frontier models had missed. A model that can find and patch flaws unaided would help an attacker just as readily, so putting it in defenders' hands first is a sensible order. Everyone else waits while Google strengthens four safeguard areas: misuse defenses that include monitoring the model's internal activations, resistance to indirect prompt injection, monitoring of its chain of thought and actions for misalignment, and hardened sandboxes for high-risk testing.
Pricing is already public. Gemini 4 Argon launches at $2 per million input tokens and $10 per million output tokens, with cached input 95% cheaper. Those are introductory rates that rise to $4 and $20 afterward, and Google has not said when the introduction ends. A team costing out an API project should plan around the higher pair.
