Gemini 4 Release Date: Argon Access & Pricing

Gemini 4 release date: Google unveiled Gemini 4 Argon on Sep 30, 2026, but only vetted cyber defenders can use it. See who gets access next and the cost.

That is the short version. Below we separate what Google has actually confirmed from the guesswork floating around X and Reddit, explain what the staged rollout means for you, and look at whether the independent benchmark numbers back up the “most powerful Gemini” headline.

Gemini 4 Argon release date: what is confirmed

Google DeepMind introduced Gemini 4 Argon on its official blog on September 30, 2026. It did not ship it to the public. The rollout order Google gave is:

• Now: a set of trusted cyber defenders through the Fairwind Program, plus early testers. Google says these partners get Argon without its usual cyber guardrails so they can find and patch vulnerabilities.
• Next: paid Gemini API customers and Google AI Ultra subscribers.
• Later: developers, enterprises and consumers more broadly, “as soon as possible”.
• No dates: Google has not given a day, week or month for any of those steps.

We checked Google’s public Gemini API model list, pricing page and changelog on October 12, 2026: none of them mention Argon yet. The newest public models there are still the Gemini 3.x family. So if a site gives you a firm Gemini 4 release date or a model ID to paste into your code, it is guessing.

Why the rollout is this slow

Google is not holding Argon back for marketing reasons. The blog post spends more space on safeguards than on features: monitoring the model’s internal activations for misuse, watching its chain of thought for misalignment, hardening the sandboxes it runs in and taking part in the U.S. government’s voluntary pre-release testing. A model that can find and patch serious security bugs on its own can also help attackers, so Google is letting defenders use it first.

Our read: this looks like a weeks-to-months staged release, not a one-day launch. For background on why labs gate their most capable agents, see our explainer are AI agents safe?. The useful signal to watch is the first appearance of an Argon model ID in Google’s API docs. That will come before the consumer Gemini app.

How to get Gemini 4 Argon access

If you are a developer

Set up a paid Gemini API account now. Google names paid API customers as the first wider group, so a free-tier key may not be enough. When access opens, the model should appear in the API’s models list for your project. Until then, there is nothing to call, and Google’s announcement does not include a public waitlist.

If you are a Gemini app user

Google AI Ultra is the consumer plan Google names for early access. But Ultra is a chat subscription, not an API key, and Google has given no Argon usage limits for it. We would not buy Ultra just for Argon until Google confirms it is switched on for subscribers.

If you work in security

The Fairwind Program is the only route today. Google describes it as a set of trusted cyber defenders, and partners such as Wiz are already using Argon. There is no open application form in the announcement, so this path is for established security teams Google already works with.

Gemini 4 Argon pricing

Google published API prices before the model is even available, which is unusual and helpful for planning:

• Introductory: $2 per million input tokens and $10 per million output tokens.
• After the intro period: $4 input and $20 output per million tokens.
• Cached input: 95% off the input price.
• Output limit: up to 1 million output tokens per response, up from 64,000.

That last point matters for your bill. Argon is built for long, multi-step jobs and can write hundreds of thousands of tokens in one run. One response that uses the full 1M-token output limit would cost $10 at the intro price and $20 afterwards, before any input. Google has not said how long the introductory period lasts, so budget at the higher rate.

Is it really the best model? What independent tests say

Google’s own numbers are strong: a state-of-the-art 77.9% on DeepSWE v1.1 for long software-engineering tasks, first place on the Vals Index of finance, legal, tax and coding work, 51.3% on Zapier’s AutomationBench, 91.7% on LVBench for long video understanding and a tie for first on CWE-bench v1 at 68%. These are vendor-reported results.

The independent picture is less clear-cut. Artificial Analysis lists Gemini 4 Argon (high) at 53 on its Intelligence Index, 8th of 227 models it tracks, and marks it as not publicly available. Among the leaders on that page, Claude Opus 5.5 scores 58, while Claude Fable 5.1 and GPT-6 Astra also score 53, and GPT-6.1 Sol 52. Where Argon stands out is cost: about $1.99 per index task, against $5.98 for Opus 5.5 and $3.26 for GPT-6 Astra.

So our take is that Argon is a top-tier model, not a clear winner on general intelligence. It looks strongest on coding agents, enterprise work and security, at a lower price than its closest rivals. For a broader ranking, see our best AI model guide and our GPT-6 Astra vs Sol comparison.

What to do while you wait

• Build on current Gemini models: the Gemini 3.x models in the API are the closest stand-in, and switching models later is usually a one-line change. Our Claude API vs Gemini API comparison covers the trade-offs.
• Test coding agents now: if coding is why you want Argon, try today’s options in our best AI agents for coding roundup.
• Watch the official channels: the Gemini API changelog and the Google blog will announce the model ID and the next access wave first.

Frequently asked questions

Sources: Google blog post “Gemini 4 Argon: our next era of frontier intelligence” (Sep 30, 2026); Gemini API models, pricing and changelog pages; Artificial Analysis Gemini 4 Argon model page. All checked October 12, 2026.