Gemini 4 Argon: Release Date, Price and How to Get Access (2026)

What Google announced on September 30

Google calls Gemini 4 Argon its "next era of frontier intelligence." In plain terms: it's Google's new top model, built for long, multi-step work. Think writing and fixing code across huge codebases, research-heavy knowledge work in finance and law, and finding and patching security holes.

The headline numbers are real, but they're specific. Argon scored 77.9% on DeepSWE v1.1, a real-world software engineering test. It scored 51.3% on Zapier's AutomationBench, where it ranks first, and 91.7% on LVBench, a long-video understanding test. Google also says it leads the Vals Index across finance, coding, legal and tax tasks.

The quiet big change is output length. Argon can write up to 1 million tokens in a single run, up from 64,000. That's the difference between a model that drafts one file and one that can work through an entire migration without stopping. Google says it used Argon internally to convert more than 800,000 lines of C and C++ to Rust, and to free over 300 TiB of memory across its data centers.

What Google did not announce: a date for the rest of us. That's what most people searching today actually want to know, so let's start there.

Who can use Gemini 4 Argon right now?

Almost nobody outside Google. The first rollout goes to "a set of trusted cyber defenders" through the Fairwind Program. Coverage of the launch describes that group as vetted organizations: national cybersecurity authorities, government agencies, and operators of critical infrastructure such as healthcare, telecom and utilities.

Why start there? Argon is unusually good at finding and fixing software vulnerabilities. Google is giving defenders a head start to patch their systems before the model is widely available, and those defenders get it without the cyber guardrails the public version will carry. Google is also taking part in the U.S. government's voluntary pre-release model access process.

Here's the rollout order from Google's own post. Only the first step has started.

If you're on the free Gemini app or Google AI Pro, there's no announced date for Argon at all. Anyone promising you "Gemini 4 Argon access" today, whether it's a browser extension, a chatbot or a "free Argon" website, is selling you something else.

How much does Gemini 4 Argon cost?

For developers, Google published API pricing up front. During the introductory period it's $2 per million input tokens and $10 per million output tokens, with cached input 95% cheaper. When the intro period ends, it doubles to $4 and $20.

For context, the intro price matches OpenAI's GPT-6.1 Sol and sits well below the flagship models from OpenAI and Anthropic.

What does a million tokens buy? Roughly 750,000 words. Feeding Argon a 60-page contract (about 40,000 tokens) would cost around 8 cents at the intro rate. A 3,000-word summary back costs about 4 cents. For one person, the model price is rarely the problem. Access is.

For consumers, there's no Argon-specific price yet. Google AI Ultra is Google's top consumer subscription and it's first in line. Google hasn't said whether cheaper plans will get Argon, or when. Check Google's own plan page before you trust any number you see in a forum.

What Argon is good at, and where it still trails

Coverage of Google's benchmark table has Argon leading or tying in 13 of the 18 tests disclosed. The biggest leads are in work that looks like a real job: long agent tasks, finance research and legal drafting.

It isn't a clean sweep. GPT-6 Astra beats Argon on FrontierSWE v2 (65.5% vs 55.0%) and Terminal-Bench Science (68.1% vs 57.6%). Claude Opus 5.5 leads on Terminal-bench 4.0 (66.4% vs 57.4%). Translation: if your work is hard, terminal-heavy engineering, the rivals still have an edge on paper.

Two honest caveats. First, these are mostly Google-reported numbers, and benchmarks have been a noisy guide this year. Second, almost nobody can test Argon yet, so there are no independent real-world reviews. Treat launch-day charts as a trailer, not the film.

One result is worth noticing if you care about AI that acts on your behalf. On Gray Swan's indirect prompt-injection test, attacks succeeded against Argon only 0.7% of the time. That test checks whether a malicious email or web page can hijack an AI agent. Lower is better, and 0.7% was the best score in the table.

Will Gemini 4 Argon come to the Gemini app, Gmail or WhatsApp?

In Google's own products, probably, eventually. Google says Argon will reach consumers "as soon as possible," and reporting expects it to end up powering most Google services the way Gemini 3 did. There are no dates yet.

WhatsApp is a different story. Google's announcement doesn't mention WhatsApp, and WhatsApp belongs to Meta, a direct Google competitor. Don't expect an official "Gemini 4 Argon in WhatsApp" button.

So before you refresh the Gemini app every hour, ask what you actually want a smarter model to do for you. For most people it isn't migrating 800,000 lines of C to Rust. It's the boring stuff: remembering the dentist, chasing the invoice, reading the contract before Thursday.

What to do while you wait: get the admin done now

A frontier model is only useful if you can reach it where you already are. If your day runs through WhatsApp, the fastest win isn't a smarter model you can't use yet. It's an assistant in the chat you already have open.

That's Agent Sonic: a personal AI assistant that lives inside WhatsApp. There's no app to install and no dashboard. You message it like any other contact.

"Remind me to renew the car insurance on the 14th at 9am." Done, and it fires in the same chat. "Every Monday at 8:30, remind me to send the weekly report." Recurring, no setup. Forward a 40-page PDF and ask "What's the cancellation clause and the notice period?" You get the answer, not a summary of a summary.

"Ask Dana when the invoice will be ready and let me know." On the Pro plan, Sonic's outreach tasks (in Beta) show you the exact first message before anything goes out. Once you say yes, Sonic messages Dana from its own number as your assistant and reports back with her answer. One task at a time, up to three people a day.

Sonic also browses the web live. Ask "Has Google opened Gemini 4 Argon to AI Ultra yet?" and you get today's answer instead of refreshing a news feed. Get access at tryagentsonic.com/contact: leave your details and you'll get Sonic's number and a quick onboarding.

How to know the moment Argon reaches you

API developers: watch the model list in Google AI Studio and the Gemini API docs. When Argon shows up for your project, you're in. Check the pricing page too, because the $2/$10 introductory rate won't last forever.

Google AI Ultra subscribers: watch the model picker in the Gemini app. Google usually announces consumer rollouts on its blog and inside the app itself.

Everyone else: set a reminder instead of checking every day. "Remind me on October 15 to check if Gemini 4 Argon is on my plan" takes five seconds to send and saves you two weeks of refreshing.

The bottom line: Gemini 4 Argon looks like a serious step up for long, complicated work, and its price undercuts the flagships. But for now it belongs to cyber defenders. Put the date question on autopilot, and spend the time getting your own admin off your plate.

Frequently asked questions