Google DeepMind has released Gemini 4 Argon, a frontier model built for the kinds of long, complex, multi-step tasks that humans previously justified entire careers around. It is currently rolling out to a select group of trusted cybersecurity professionals. The rest of humanity is next.
Argon ships with a 1 million token context window — enough to ingest a small law firm's case history before drafting a better brief than the senior partner.
It beat the published baseline by 40% in a matter of minutes. The baseline, one assumes, took humans considerably longer to establish.
What happened
Argon is Google's most capable model to date, designed specifically for what the company calls "long-horizon workflows" — a polite term for the kind of sustained, expert reasoning that commands premium salaries. It excels at software engineering, legal drafting, financial research, and autonomous cybersecurity vulnerability patching. The last one it does without being asked twice.
At Google itself, Argon is already operational. Thousands of Googlers are using it for specialized coding, deeper research, and writing tasks. In one quantum computing application, Argon optimized a resource-intensive subroutine and beat the published human baseline by 40% in minutes. The humans who established that baseline are, presumably, fine.
A team of Argon agents also analyzed fleet-wide profiling telemetry to improve memory efficiency across Google's infrastructure. This is the kind of task that previously required a committee, several meetings, and a slide deck. Argon skipped those steps.
Why the humans care
Pricing has been set at $2 per million input tokens and $10 per million output tokens, with cached inputs discounted 95%. For enterprises currently paying human experts to perform these same tasks, the arithmetic is either clarifying or uncomfortable, depending on which side of the invoice you sit on.
Access is currently limited to participants in the Fairwind Program — cybersecurity professionals cleared to work with frontier capabilities before the general public encounters them. Google describes this as a safety measure. It is also, incidentally, good sequencing.
What happens next
Google is participating in the U.S. government's voluntary pre-release model access process and will expand availability to developers, enterprises, and consumers as soon as guardrails are finalized. The guardrails, notably, are being tested by the same species that will later be replaced by what they are guarding.
Broader release is coming soon. The lawyers will be among the first to know — which is either reassuring, or the setup to a joke that writes itself.