Google announced Gemini 4 Argon, its new frontier AI model, on September 30, 2026. You cannot use it yet: for now it is going only to a group of trusted cyber defenders. Paid API customers and Google AI Ultra subscribers come next. Google has not given a date.
What we know
- What it is: Google's new frontier model. Google presents it for long, complex work in three areas: software engineering, specialized business work such as legal and finance, and cybersecurity defense.
- Who has it today: a group of trusted cyber defenders, through Google's Fairwind Program.
- Who gets it next: developers, enterprises and consumers, starting with paid API customers and Google AI Ultra subscribers. Google says as soon as possible and gives no date.
- Developer pricing: an introductory $2 per million input tokens and $10 per million output tokens. After the introductory period, $4 and $20.
- Longer answers: the output limit goes from 64,000 tokens to 1 million.
- Security: Google says the model can find, validate and patch critical software vulnerabilities on its own. It says capabilities at this level call for a phased release, and that it is taking part in the US government's voluntary process for pre-release model access.
- Performance figures: the announcement cites several benchmarks, including 77.9% on DeepSWE v1.1. These are Google's numbers, and the model is not open for anyone to check them.
- Context: OpenAI sells GPT-6.1 Sol, introduced in September, at the same API price: $2 and $10 per million tokens.
What changes and what doesn't
Nothing changes today if you use Gemini on your phone or at work. The announcement does not say when the model reaches each product.
It is not an SEO change either. Google's post says nothing about Search or about how results are ranked.
Two things change if you pay for AI. One is price: two of the latest models from Google and OpenAI come in at the same cost per token for as long as Google's introductory price lasts. The other is the size of the answer: with an output limit of 1 million tokens, a model can hand over much longer work in one go.
And there is a warning if you run a website. If a model can find security flaws on its own, keeping your CMS and its modules up to date matters more than before.
What you can do today
- Do not buy anything because of this announcement. There is nothing to buy yet.
- If you pay for an AI API, write down both prices and budget with the regular one, not the introductory one.
- If you pay for Gemini, check which plan you are on. Google names AI Ultra subscribers first.
- Prepare a test of your own: three or four real tasks from your business with the result you expect. With them you can compare any new model in an afternoon.
- Update your website and change the passwords nobody has touched in years.
How we apply it at DomHostSeo
We use AI to write and illustrate, and we check every fact against its primary source before publishing. We have not tested Gemini 4 Argon: it is not open to us. So here we only report what Google's announcement says, which we read on October 2 and 3, 2026, and we flag that the figures are Google's.
That check applies whichever model does the writing. We covered it in Google: Fact-Check AI Content by Hand Before Publishing.
Sources
Updates: we will add the opening date for the API and for Google AI Ultra here when Google announces it.