Gemini 4 Argon Explained: You Cannot Use It Yet, and Ultra Gets It First

Google announced Gemini 4 Argon on September 30, 2026, and the short answer for anyone who is not a professional cyber defender is that you cannot use it and Google has not said when you will. Argon is a frontier model aimed at long, multi-step professional work: software engineering, legal and financial research, and finding and patching security holes. Google is releasing it first to a small vetted group, and when it does reach ordinary paying customers, the announcement names exactly one consumer plan.

Key Takeaways

• You cannot use it today. Google says Argon "is rolling out to a set of trusted cyber defenders through our Fairwind Program." That is the whole of its availability.
• Google AI Ultra is first in line. Verbatim: Google will "release to developers, enterprises, and consumers, starting with paid API customers and Google AI Ultra subscribers." Ultra costs $99.99 a month and is billed monthly only.
• There is no date. Google says "as soon as possible," after it finishes "strengthening critical frontier safeguards."
• Your Gemini app did not change. Google's plan page, re-read October 1, 2026, still lists the app models as Gemini 3 Flash-lite, Gemini 3 Flash and Gemini 3 Pro, all available on every plan including the free one.
• No Google surface names Argon except the blog post. The Gemini plan matrix, gemini.google/subscriptions/, one.google.com/about/google-ai-plans/ and deepmind.google/models/gemini/ all contain zero mentions, read October 1, 2026.
• The published prices are developer prices. $2 per million input tokens and $10 per million output tokens introductory, rising to $4 and $20. There is no consumer price because there is no consumer product.
• The reason for the delay is stated, not hidden. Google says it is in "the U.S. government's voluntary process for pre-release model access" and is hardening the model against misuse and prompt injection before a wider release.

Can You Use Gemini 4 Argon?

No, and not by paying. Google's announcement gives Argon exactly one availability sentence: it "is rolling out to a set of trusted cyber defenders through our Fairwind Program." Fairwind is Google's vetted-access scheme for its security models, which it first used for Gemini 3.8 Flash Cyber and which admits government authorities, critical infrastructure operators and software maintainers rather than individuals. We covered how that gate works when it appeared in Gemini 3.8 Flash.

There is no waitlist, no toggle, and nothing to subscribe to. Google is explicit that the broad release has not happened: "We'll continue to gather feedback from early testers as we iterate on guardrails before making Argon available to developers, enterprises, and consumers as soon as possible."

Who Gets It First, and What It Costs

When Argon does open up, Google has published the order. This is the one sentence in the announcement that answers a consumer question, and it names a plan:

"We're grateful for the initial cohort of cyber defenders and trusted testers whose real-world evaluations and feedback will help us strengthen our systems before we release to developers, enterprises, and consumers, starting with paid API customers and Google AI Ultra subscribers."

Plan prices read from one.google.com/about/google-ai-plans on October 1, 2026. That page has no free column, and its "Access to latest models" row is the relevant one: it reads Limited for AI Plus, Expanded for Pro, Higher and Highest for the two Ultra tiers. Google does not promise anywhere that a new frontier model reaches Pro or the free tier at all, and with Argon it has now said in writing that Ultra comes first.

The only numbers Google has published are developer prices. Argon launches at an "introductory price of $2 per million input tokens and $10 per million output tokens, with cached input tokens priced at 95% off input token price." The footnote adds what happens next: "After the introductory period expires, the price of $4 per 1M input tokens and $20 per 1M output tokens will apply." None of that is a subscription, and none of it is something you can buy in the Gemini app.

What Argon Is Actually Built For

Long jobs, not chat. Google's framing is "deep reasoning across complex, long-horizon workflows," and the concrete change behind it is the output limit: Google says it expanded the model's output token limit to 1M tokens, up from the previous 64K. In plain terms, the model can produce a very long piece of work in one go instead of stopping and being restarted, which is what a multi-hour task needs and what a conversation does not.

The benchmark results Google published, all from the announcement:

Google also published what Argon has done inside Google, which is the clearest signal of what the model is for. It says Argon agents freed "over 300 TiB of memory" across Google's data centres with an estimated 500 TiB to 1 PiB in total, are migrating C and C++ code to Rust "up to 800K+ lines for the Fuchsia Zircon kernel," and rewrote 32,000 lines of hand-tuned code in Google's open-source video decoder into memory-safe Rust that "runs 2.7x faster than the Rust port." None of this is a feature you would ever see in a chat box.

Why Google Is Holding It Back

Google says why, in unusual detail, and the reasons are about security rather than polish. It names four areas it is still strengthening before "broad availability": refusing requests that could help with cyber or chemical, biological, radiological and nuclear attacks; resistance to prompt injection, where hidden instructions in a document or web page hijack the model; monitoring for the model pursuing a task in ways the user did not intend; and sealing off the sandboxes it is tested in.

It also says it is "actively engaged in the U.S. government's voluntary process for pre-release model access while we gradually expand access."

The security capability is the reason both for the early release and the delay. Google trained Argon to "autonomously find, validate, and patch critical software vulnerabilities," and for Fairwind defenders it is "releasing Argon without cyber guardrails." It cites a live result: the security firm Wiz used Argon and "uncovered a critical vulnerability exposing sensitive personal information across healthcare software used by hospitals worldwide, identifying a severe risk that previous frontier models had missed." A model that good at finding holes is a model Google does not want handed to everyone on day one.

What To Use Instead Right Now

Use what is in your app, because nothing about it changed on September 30. Google's Gemini plan page, read October 1, 2026, lists three models for the Gemini app -- Gemini 3 Flash-lite, Gemini 3 Flash and Gemini 3 Pro -- and marks all three available on every plan, including "Without an AI Plan." What the plans actually change is how much you get:

For the question underneath all of this -- whether a $19.99 Gemini plan or a $20 ChatGPT plan is the better buy for an ordinary week -- our which AI chatbot should you pay for comparison works through it with current prices, and Gemini 3.1 Pro covers the model a paying Gemini subscriber is actually talking to. If you are weighing the free tier instead, best free AI chatbot 2026 is the one to read.

Should You Pay For Ultra To Get Argon?

No, not yet, and not for this. Google has named Ultra as first in line and named no date. Paying $99.99 a month today buys you higher limits on Gemini 3 Pro, 20 TB of storage and the other Ultra bundles, not Argon. Nothing Google has published commits it to a month, a quarter or a year.

Wait if you were going to upgrade anyway. The announcement is a reason to expect Ultra to keep getting new models first, which is useful information for a decision in a few months, not a reason to upgrade this week.

Ignore it entirely if you use Gemini for ordinary things: writing, summarising, images, questions. Argon is built for eight-hour engineering jobs and vulnerability hunting. On the evidence Google published, it would make no difference to a normal chat even if you had it.

The honest summary: Gemini 4 Argon is a real and significant model, it is the clearest statement yet that Google AI Ultra is where new Gemini models land first, and for now it has nothing to do with the app on your phone.