Google's Most Powerful AI Model Went to Cyber Defenders First. That's the New B2B Playbook.

Introduction of Gemini 4 Argon

The most capable AI model Google has ever built launched today, and almost nobody can use it. Gemini 4 Argon, announced September 30 by Google DeepMind, is rolling out first through Google's Fairwind Program to what the company calls "trusted cyber defenders." Paid API customers and Google AI Ultra subscribers are next in line. Everyone else waits.

What Google actually shipped

Koray Kavukcuoglu, SVP of Google DeepMind and Chief AI Architect at Google, framed Argon around three jobs: real-world software engineering, enterprise knowledge work like legal and finance, and cybersecurity defense.

The headline numbers:

• 1 million output tokens, up from 64K. The model can produce an entire codebase migration, a full financial analysis, or a complete legal draft in a single pass.
• 51.3% on AutomationBench, ranked first, a benchmark built around executing real business processes.
• 77.9% on DeepSWE v1.1 for real-world software engineering.
• 68% on CWE-bench v1 for vulnerability remediation, tied for first.
• Leading scores on the Vals Index across finance, coding, legal, and tax.

Introductory API pricing is set at $2 per million input tokens and $10 per million output tokens, with a 95% discount on cached input. Standard pricing of $4 and $20 applies after the introductory window.

Why cyber defenders got the keys first

Fairwind launched on September 2 and already includes more than 650 organizations: government agencies, critical infrastructure operators in healthcare, telecom, energy, and finance, plus security partners like CrowdStrike, Palo Alto Networks, Snowflake, and Wiz. Participants agree to strict operating standards, including limiting access to internal security and incident response teams.

Early results back the approach. Wiz, through its Scan for Good program, used Argon to surface a critical vulnerability in healthcare software that previous frontier models had missed.

The B2B takeaway: access is the new differentiator

Here's what every enterprise leader should take from this launch.

• Trust is now a procurement requirement. The organizations that got Argon first had already shown they could handle powerful tools responsibly. Expect more AI vendors to tier access by security posture, governance maturity, and use case. Your AI governance program is no longer a compliance line item. It's your ticket in.
• The target is knowledge work, not chat. AutomationBench, legal agent benchmarks, finance, tax. Google built Argon to do the work that fills a professional services invoice. The competition among frontier labs has moved from "who has the best chatbot" to "who can run your business processes."
• Output length changes the unit of work. A 1 million token output ceiling means the deliverable is no longer a paragraph you stitch together. It's the whole thing. Teams that redesign workflows around complete, end-to-end outputs will move faster than teams bolting AI onto old processes.

• No model wins everything. Argon leads on knowledge work and several engineering benchmarks, but enterprise buyers should still evaluate models per workload, not pick one vendor and hope.
• Google is its own proof point. Thousands of Googlers already use Argon internally. It freed more than 300 TiB of data center memory, handled an 800,000-line kernel rewrite, and made a video decoder 2.7x faster. When a vendor runs its own infrastructure on the product, that's the case study that matters.

What to do this quarter

• Audit your AI governance posture. If gated access becomes the norm, your security and data controls determine what you can buy.
• Map your highest-cost knowledge workflows: contract review, financial modeling, codebase modernization. Those are the workloads this generation of models is built for.
• Build a model evaluation process by task, not by brand.
• Get on the API waitlist now if you're a paid Google Cloud or API customer.

The bottom line

Gemini 4 Argon isn't just a bigger model. It's a signal of how frontier AI will reach business from here on: defenders first, trusted partners next, everyone else once the guardrails hold. The companies that prepare for that model of access will be the ones using the most powerful tools first.