Google names Gemini 4 Argon its new frontier model
Google announced Gemini 4 Argon, its next frontier model, in a Gemini 4 Argon announcement from Google DeepMind written by Koray Kavukcuoglu, the SVP of Google DeepMind and its Chief AI Architect. This is a distinct model, a step above the Gemini 3.8 family that already runs inside the consumer app and Google's AI Mode in Search.
Google points the model at the work partners actually care about: software engineering, and enterprise knowledge tasks it names directly, including finance, legal research and tax. It also calls out cybersecurity defence, where the model can patch software vulnerabilities on its own. Google reports frontier-level results on its own internal evaluations, which are the maker's figures and not independent tests. The headline practical change is capacity: the output limit rises to one million tokens, up from the previous 64,000, with the input limit at one million as well. In plain terms, it can read and write far longer documents in a single pass. Google's own September 2026 AI recap lists the release alongside the month's other model news.
For now, almost no one can switch it on
Here is the catch. Access is tightly gated. Today the model is open only to trusted cyber defenders through Google's Fairwind Program. Google says developers, enterprises and consumers will follow "as soon as possible," with paid API customers and Google AI Ultra subscribers named as the next in line. It has not given a general-availability date. So this is a confirmed launch, but a staged one, and the version most firms could reach is not out yet.
A capability story, not a to-do list
For a professional-services firm that wants to be found, there is nothing to action here and nothing to switch on. That is the calm read: a flagship release worth knowing about, with no local lever to pull today. The quiet signal underneath it is that the assistants your prospects use are getting more capable at reading, reasoning over and citing the web. The firms that benefit are the ones whose public facts are accurate and easy to quote, which is the heart of earning recommendations from AI assistants rather than hoping to be guessed at. If finance, legal or tax work is your world, it is worth watching how these tools describe your field, a practical concern for accounting and advisory firms as much as anyone. When you are ready to turn that into a plan, that is what AI-visibility work is for.
Frequently asked questions
Can my firm use Gemini 4 Argon now that it has launched?
Gemini 4 Argon is not open to most firms yet. Access is limited to trusted cyber defenders inside Google's Fairwind Program. Google says developers, enterprises and consumers will follow "as soon as possible," with paid API customers and Google AI Ultra subscribers named next, but it has not published a general-availability date.
Does the Gemini 4 Argon release change how my firm gets found in AI answers?
The Gemini 4 Argon release does not change what your firm should do to be found. There is nothing to configure or switch on. This is a capability release, and for most firms the only sensible step is to keep the public facts about your business accurate and easy to quote, because more capable assistants read and cite the web.
What does Gemini 4 Argon cost for developers to use?
Gemini 4 Argon's introductory API pricing is $2 per million input tokens and $10 per million output tokens, with cached input 95% cheaper; standard pricing then rises to $4 and $20. Those are Google's published figures, and they apply to developers building on the model, not to everyday chat users.
Is Gemini 4 Argon the same as the Gemini already in Google's app?
Gemini 4 Argon is not the same as the Gemini already in the app. It is a new frontier model, separate from the Gemini 3.8 family that powers today's consumer app and Search features. Google positions it above them as its most capable model.
