GEMINI 4 ARGON · AVAILABILITY • 13 min read •

Updated

Gemini 4 Argon release date: who had it first, and who has it now

The Gemini 4 Argon release date was 13 October 2026, when it became generally available. In the Gemini API it runs on the paid tier only. A million input tokens cost $2 there and a million output tokens $10 until 30 November. From 1 December you pay $4 and $20. The same day it started rolling out to Google AI Pro and Ultra subscribers in the Gemini app, worldwide. Before that, from 30 September, it was rolling out to some Fairwind Program partners. Trusted testers and the US government had it before the release too, but Google does not say since when.

A rope barrier before a closed door with green light spilling out

Google announced Argon on 30 September at 20:00 UTC (22:00 in Poland). For two weeks it gave no date and said only: as soon as possible. The release came on 13 October. I wrote the first version of this piece before it. On 23 October I updated it from Google's price list, changelog and documentation. I have not tested the model.

The timeline, the 1M-token limit on a single answer and the test results are in my map of the whole Argon announcement, from benchmarks to Google's internal examples. Here I stay with one question. Who got the model, and when?

When did Gemini 4 Argon become available?

Since 13 October 2026. That day the Gemini API changelog announced the generally available (GA) release of gemini-4-argon. Google Cloud gives the same release date. Before that, Google wanted to gather feedback from early testers and iterate on guardrails. Reuters reported that Google did not provide a timeline for the public release. It did describe the order from the start.

Who got Gemini 4 Argon, and when (as of 23 October 2026)
StageWhoWhen
First Some Fairwind Program partners ("a set of trusted cyber defenders"), trusted testers and Google's own teams Fairwind from 30 September 2026. Testers and Google's teams earlier, on dates Google does not give
In parallel The US government, through its voluntary process for pre-release model access Google does not say
Release The Gemini API (paid tier only), Google AI Studio and Google Cloud. In the Gemini app, Google AI Pro and Ultra subscribers worldwide 13 October 2026

Before the release the model was already at work inside Google. The announcement says thousands of Googlers used it.

On the evening of 30 September Argon was not on the Gemini API pricing page or in its changelog. It was not in Google Cloud or in the Gemini app release notes either. I took a changelog entry with the model ID as the launch signal. It appeared on 13 October. The same day the Gemini app release notes announced the model for the AI Pro and Ultra plans.

What is the Fairwind Program, and who gets Argon without guardrails?

Fairwind is a limited-access program for governments and trusted partners. Google launched it on 2 September with Gemini 3.8 Flash Cyber and the CodeMender tool. It has more than 650 partners worldwide. Before the release only some of them got Argon. Google wrote of "a set of" partners with exclusive access. How many? It did not say. On 13 October Google began expanding access to hundreds more Fairwind partners.

Google named only one company that got Argon before the release. It is Wiz, which uses Argon in its Scan for Good initiative to find and fix high-risk exposures in public infrastructure for free. Wiz has belonged to Google since March 2026. The Fairwind pages also show partner logos, CrowdStrike and Palo Alto Networks among them. The program started with 3.8 Flash Cyber, and Google did not say those companies got Argon.

Who can apply? Google lists three groups: governments and national cyber authorities, critical infrastructure operators from healthcare to financial networks, and core technology platforms. It runs background checks on every organization that applies. Inside an accepted organization, only internal cybersecurity, incident response and penetration testing teams may use Argon. Sharing or selling access is not allowed.

One sentence raised the most questions. Google said trusted defenders and its own teams would get Argon "without cyber guardrails", so they could use its full defensive capabilities. That version now has a name: Gemini 4 Argon Cyber Permissive. On Google Cloud it is available only behind an allowlist, through the Fairwind Program or your Google team. It runs only in the global and us regions. Access to 3.8 Flash Cyber does not carry over to it, so you have to apply separately. Even Argon without guardrails has limits. Partners may run authorized threat simulation, reverse engineering and malware analysis, for defensive and academic research. Creating malware is not permitted.

Here Google settles at the model level what every company has to settle in-house: who gets full capability, for what, and under whose oversight. Inside a company, the question is about AI agents. I have written up how to give each AI agent a scope of autonomy and a threshold at which it asks a human.

The release post invites defenders to apply to the program for Argon without cyber guardrails, and Google admits them on a rolling basis. The application form is on the program page. Google promises a reply to organizations that meet its criteria. It points the rest to CodeMender with publicly available models.

Who in the US government had Argon before release?

Google does not say. It writes only that it is "actively engaged in the U.S. government's voluntary process for pre-release model access". Sundar Pichai put it more briefly on X: "it's with the US gov't". Nor is it known how long the review lasted or what it found. Reuters calls it the Trump administration's voluntary process.

A US executive order of 2 June 2026 calls for a voluntary framework of this kind. Under it, a developer could give the federal government access to a model for up to 30 days before releasing it to other trusted partners. Google does not say whether that is the process it means.

This matters for companies outside the US. The example comes from Anthropic. On 12 June the US government issued an export control directive that suspended access to Claude Fable 5 and Claude Mythos 5 for any foreign national. Anthropic had to disable both models for all its customers. The controls were lifted on 30 June. None of this involves Argon. It does show that for models strong in cybersecurity, the US government can have the last word.

In the Gemini API Argon runs on the paid tier only. The pricing page says plainly that it is not available on the Free tier. The 30 September announcement hinted at this with the word "paid". The model ID is gemini-4-argon, and it is in Google AI Studio too. On Google Cloud, in Agent Platform (formerly Vertex AI), it has been GA since 13 October.

The Standard price per million tokens is $2 for input and $10 for output until 30 November 2026. From 1 December 2026 it is $4 and $20. Thinking tokens count as output. A footnote in the release post says this price applies to the publicly available Argon model only. I have worked out separately how an Argon budget changes on 1 December, from caching to the Batch API.

In the Gemini app Argon has been rolling out since 13 October to Google AI Pro and Ultra subscribers, worldwide. That is what the app release notes say. The 30 September announcement named only Ultra. In Poland AI Pro costs PLN 97.99 a month, and Ultra PLN 469.99 or PLN 979.99. In the US that is $19.99, $99.99 and $199.99. Ultra gives the highest access to Argon. Its overall Gemini app usage limits are five or twenty times those of AI Pro, depending on the tier. The limits page gives no separate figures for Argon. The plans page does not list Argon in the cheaper AI Plus plan. For Argon alone, AI Pro is enough.

FOR AI ARCHITECTS AND CONSULTANTS

Designing or deploying agents for clients? In our collective of AI consultants you prepare for AI vendor partner certifications and learn with other consultants. When a suitable client brief arrives, we may invite you to a project.

Is Gemini 4 Argon the same as Gemini 4 Pro?

Google does not use that name. The model is officially called Gemini 4 Argon. A Google spokesperson told Reuters that Argon is larger than Google's previous line of top-tier "Pro" models. I did not find that sentence on Google's own pages.

The Gemini 4 Pro name came from leaks. On 23 September TestingCatalog reported tests of "Gemini 4 Pro" under the codename argon. On 25 September NokiaPowerUser reported a model on Arena under the label gemini-3.8-flash, although that is also the name of a public Google model. The leaks spoke of a 256K-token output limit. The official figure is 1M. Apart from the name Argon, Google confirmed none of these reports.

Why did people expect a Pro model? According to The New Stack, Google announced plans for Gemini 3.5 Pro in May, and a series of Flash models came out instead. The New Stack calls Argon its replacement. So if you are looking for a Gemini 4 Pro release date, Argon came out on 13 October. Google has announced no product called Gemini 4 Pro.

Is Gemini 4 Argon available in Poland and the EU?

According to Google's pages, yes. The Gemini API and Google AI Studio are available in all 27 EU countries, and the list of available countries makes no exception for Argon. The Gemini app offers it to AI Pro and Ultra subscribers worldwide, and both plans are on sale in Poland. I found no separate announcement for Poland or the EU.

Before the release I saw two patterns. New models usually started everywhere at once. Gemini 3 Pro rolled out "to everyone" in the Gemini app, 3.1 Pro "globally", and the API changelog gave no regional limits for either. New features for paid plans often skip Europe, though. On 1 October Gemini Spark, an AI agent for AI Pro and Ultra subscribers, worked everywhere except the EEA, Nigeria, Switzerland and the UK. Google has since renamed it Task, and the excluded countries are the same. Personal Intelligence launched only in the US. Argon followed the first pattern.

OpenAI showed the second pattern the day before Argon was announced. I described how the Pro plan got ChatGPT's always-on agents everywhere except the EEA, Switzerland and the UK. Agent features from American companies often start in the US.

Then there is where your data is processed. On Google Cloud, pay-as-you-go use of Argon (Standard PayGo) works on the global endpoint and in the us and eu multi-regions. The europe-west2 and asia-southeast1 regions support only reserved capacity (Provisioned Throughput). europe-west2 is London, so it is outside the EU. For comparison, on 1 October Gemini 3.1 Pro Preview ran there only on the global endpoint. 3.8 Flash Cyber had a global and a US endpoint, and 3.8 Flash an EU region as well.

Before the release I guessed that the API would open in Poland on the same day as in the US, as it did for Gemini 3 Pro and 3.1 Pro. That is what happened. The eu multi-region on Google Cloud, which I expected later, is already there too.

What to do this week

The model is out. I would do six things.

  1. Test Argon on your own tasks. Pick a few dozen real tasks and write down what a good result looks like. Run Argon and your current model on them.
  2. Budget at $4 and $20. The $2 and $10 price holds until 30 November. From 1 December you pay twice as much.
  3. Check the region. If a client needs data processed in the EU, choose the eu multi-region on Google Cloud. The europe-west2 region is outside the EU.
  4. Turn on billing in your API project. Argon is not available on the Gemini API Free tier.
  5. Apply to Fairwind if you defend critical infrastructure. You get the Cyber Permissive version through that program or through your Google team.
  6. Keep a second model. The Fable 5 episode showed that a model can disappear overnight.

Start with the task list. Open a spreadsheet today and write down the first five.

Newsletter

The Agentic Architect

Practical patterns, case studies, agentic projects taken apart and AI news. Zero spam, once a week.

You'll get one email to confirm. Unsubscribe with one click. Privacy policy

Sources and fact-check date

First published 1 October 2026. Updated from Google's pages on 23 October 2026.

Frequently asked questions

When is the Gemini 4 Argon release date?

Gemini 4 Argon has been generally available since 13 October 2026. It runs in the paid Gemini API, in Google AI Studio, on Google Cloud and in the Gemini app for Google AI Pro and Ultra subscribers. Before that, from 30 September, some Fairwind Program partners had it.

How much does Gemini 4 Argon cost in the API?

Until 30 November 2026 a million input tokens cost $2 and a million output tokens $10. From 1 December 2026 it is $4 and $20. Thinking tokens count as output. The model is not available on the Gemini API Free tier.

Is Gemini 4 Argon the same as Gemini 4 Pro?

Google does not use the name Gemini 4 Pro. The model is called Gemini 4 Argon. A Google spokesperson told Reuters that Argon is larger than Google's previous line of Pro models. The Gemini 4 Pro name comes from leaks before the launch.

Can I use Gemini 4 Argon in Poland?

Yes. The Gemini API and Google AI Studio work in Poland, and Argon is available there on the paid API tier. In the Gemini app Google AI Pro and Ultra subscribers have it, and both plans are on sale in Poland. On Google Cloud the model also runs in the eu multi-region.

What is the Fairwind Program?

It is Google's limited-access program for governments, critical infrastructure operators and core technology platforms. It has more than 650 partners worldwide. Some of them get Gemini 4 Argon Cyber Permissive, the version without cyber guardrails, for defensive work.

SP

Szymon Paluch

ex-CTO · AI Strategy

Planning a client rollout around a brand-new model?

The Certified AI Consultant Program is a collective of AI consultants. Together we prepare for AI vendor partner certifications. Today those are the Claude certifications, through our partner organization. The program does not grant a Google certification. You learn alongside other consultants, and after a separate profile assessment you may be invited to a client project. Applying is free; you pay only after acceptance.

See the program for AI consultants
Related posts