Gemini 4 Argon Explained: 1M-Token Output, Pricing, and When You Can Use It

Google has announced Gemini 4 Argon, the first model in its Gemini 4 generation. Revealed on September 30, 2026, it's built for long, complex work like software engineering, legal and financial tasks, and cybersecurity defense, and Google calls it a new frontier model for professional-grade work.

The catch is that almost nobody can use it yet. Argon is currently limited to a small group of vetted cyber defenders, with a wider release promised "soon" but no date. Here's what's confirmed, what Google is claiming, and what to expect next.

The Headline Feature: A 1 Million Token Output Limit

The biggest spec in Google's announcement is a 1 million token output limit, up from 64,000 in earlier Gemini models. That number describes how much the model can write in a single response, not how much it can read. Google says the extra room lets Argon think through a hard problem at length and produce hundreds of thousands of tokens in one go, rather than breaking the work into pieces.

Worth noting: Google hasn't published the model's input context window yet, so figures you may see elsewhere for how much it can take in are not confirmed.

Who Can Use Gemini 4 Argon Right Now

Only a limited group. Argon is rolling out to trusted cyber defenders through Google's Fairwind Program, the gated defense initiative the company launched in early September. Google says those defenders and its own internal teams get the model without its usual cybersecurity guardrails, so they can use its full defensive capabilities.

For everyone else, Google says broader access will begin with paid API customers and Google AI Ultra subscribers, but it hasn't said when. The company is taking part in the U.S. government's voluntary pre-release testing process and says it's still strengthening safeguards around misuse, prompt injection, and misalignment before a public launch. As of early October, Argon isn't in the public Gemini API model list.

Announced Pricing

Google says Argon will launch at an introductory price of $2 per million input tokens and $10 per million output tokens, with cached input tokens 95% cheaper than the standard input rate. After the introductory period, the price rises to $4 and $20.

That introductory rate matches what OpenAI charges for GPT-6 Sol, so the top labs are now competing closely on price. Since the model isn't generally available, treat these numbers as announced rather than final, and check Google's API pricing page when access opens.

What Google Says It Can Do

Google reports state-of-the-art results on several benchmarks. That includes 77.9% on DeepSWE v1.1, which tests long software engineering tasks, first place on Zapier's AutomationBench at 51.3%, and 91.7% on LVBench for long-video understanding. It also says Argon leads the Vals Index, which covers finance, coding, legal, and tax work, and ties for first on the CWE-bench v1 security benchmark at 68%.

Two examples from the announcement stand out. Google says teams used Argon agents to find memory optimizations across its data centers that free up more than 300 TiB once rolled out. And security firm Wiz says the model uncovered a critical vulnerability in healthcare software used by hospitals worldwide that earlier frontier models had missed.

As impressive as that sounds, every one of these figures was published by Google or its partners. With access this restricted, few outsiders have had a chance to reproduce the results, so independent testing is the thing to watch for next.

What It Means for Document-Heavy Work

Google specifically points to legal drafting, financial research, and the ability to take action based on a series of documents, plus professional chart analysis. In practice, that means long PDFs, contracts, filings, and spreadsheets, which is exactly the kind of material that benefits from some preparation. An oversized or badly scanned file wastes space and money with any AI model, and the cleaner the input, the better the output.

Compressing a file, pulling out just the pages you need, or converting a PDF into editable text before handing it to an AI tool makes a real difference. Our PDF Compressor, Split PDF, and PDF to Word tools handle that prep in your browser, and developers can tidy API payloads with our JSON Beautifier.

What to Do Until Argon Opens Up

If you're a developer, watch the Gemini API model list and Google's announcements, since a public model ID will appear there first. If you're a regular user, there's nothing to switch on yet. The sensible move is to keep using your current Gemini model and wait for independent reviews before deciding whether Argon is worth changing for.

We're also tracking how it stacks up against other big releases this quarter, including Alibaba's Qwen3.8-Max, and we'll update this story as soon as Google confirms a public release, in our AI tools and updates section.