Gemini 4 Argon Is Official. Here's Who Can Use It, and When.

Gemini 4 now has a name.

Google announced Gemini 4 Argon on September 30, 2026 US time, which was October 1 in Sydney. It is the first officially announced Gemini 4 model.

But this is not a normal public launch.

Google says Argon is rolling out first to a limited group of trusted cyber defenders through its Fairwind Program while the company continues safety work and gathers early feedback. Broader access for developers, enterprises and consumers is still to come.

Google's Argon announcement also fills in several blanks that were still open when this article was first published on September 29.

We now have:

• an official model name
• announced launch pricing
• a 1-million-token output limit
• a substantial Google-reported benchmark table
• more detail on cybersecurity and safeguards

What we still do not have is a general Gemini API model ID, a broad availability date or a published input-context specification.

Quick answer: Gemini 4 Argon is official, but most people still cannot use it. Google has begun a limited rollout and says paid API customers and Google AI Ultra subscribers will be among the first groups to receive broader access. No date has been announced.

Gemini 4 Argon at a glance

Is Gemini 4 actually out?

Yes and no.

Google has announced Gemini 4 Argon and says it is already rolling out to an initial group of trusted cyber defenders through the Fairwind Program.

That is a real rollout.

It is not general availability.

There is currently no public Argon model ID in Google's Gemini API model directory, and the ordinary Gemini API pricing page does not yet list an active Argon model entry.

Google's own wording is deliberately staged.

It says safely releasing frontier capabilities at this level requires a phased approach. The company is also participating in the US government's voluntary process for pre-release model access while it expands availability.

For most developers and Gemini users, the practical answer is therefore still:

You cannot use Gemini 4 Argon yet.

When will everyone else get Gemini 4 Argon?

Google has not published a date.

The company says it plans to make Argon available to developers, enterprises and consumers as soon as possible after the early testing phase.

It also gives us the beginning of the rollout order: paid API customers and Google AI Ultra subscribers are expected to be among the first broader-access groups.

What Google has not announced is when that happens.

There is currently no firm date for:

• public Gemini API access
• Gemini app availability
• Google AI Studio access
• broader enterprise access
• the start of Google AI Ultra availability

That means the old "Gemini 4 release date" question has changed, rather than disappeared.

The model has been announced.

The date most people can actually use it is still unknown.

Gemini 4 Argon pricing

One major unknown is no longer unknown.

Google has published launch pricing for Argon even though broad API access is not yet available.

The company says Argon will launch at an introductory rate of:

Google also states that after the introductory period the prices will become:

At the introductory input rate, a 95% cached-input discount works out to about $0.10 per million cached input tokens. At the later $4 input rate, it would be about $0.20 per million.

Those two cached figures are derived from Google's stated discount rather than separately published prices.

There are still important gaps.

Google has not stated, at publication time:

• when the introductory period ends
• batch pricing
• Flex pricing
• Priority pricing
• cache-storage pricing
• any separate long-context price tier

Those should not be borrowed from Gemini 3.x pricing.

The 1M-token number is an output limit, not a context window

This one is easy to misread.

Google says it is expanding Gemini 4 Argon's output-token limit to 1 million tokens, up from 64K.

That is an extraordinary amount of possible output.

It is not the same thing as saying Argon has a 1-million-token input context window.

Google has not published a general input-context specification for Argon.

The company's benchmark table does include a GraphWalks long-context evaluation with a 256K-to-1M test range.

That demonstrates long-context capability under that evaluation.

It does not automatically define the production API's input-context limit.

Until Google publishes a product specification, the accurate wording is:

Argon has a stated 1M-token maximum output. Its public input-context limit has not been specified.

Google's benchmarks are strong, but Argon does not win everything

Google has now published a substantial Gemini 4 Argon evaluation table.

These are Google-reported results, not Choosely hands-on testing or a fully independent leaderboard.

A useful selection is below.

Google's current Gemini 4 Argon performance page contains the full table and methodology link.

The broad pattern is more useful than declaring a winner.

Argon performs particularly strongly in Google's reported knowledge-work, long-video, science and several coding evaluations.

But GPT-6 Astra remains ahead on FrontierSWE v2, Terminal-Bench Science 0.1 and OSWorld 2.0 in Google's table. Claude Opus 5.5 leads Terminal-bench 4.0 and PostTrainBench.

That matters.

Google's launch data makes Argon look highly competitive.

It does not establish that Argon is simply better at everything.

A benchmark table is still not the product

The comparison also has an important limitation.

Agent benchmarks often measure more than a bare model.

Different labs may use different harnesses, tools, inference configurations and system setups.

Google publishes methodology notes for its Argon evaluations, and those details matter before turning small percentage differences into sweeping model rankings.

Choosely will treat these figures as launch evidence until Argon is broadly available for reproducible real-world evaluation.

Why cybersecurity is part of the limited rollout

Google is giving Argon to cyber defenders early for a reason.

The company says Argon can autonomously find, validate and patch software vulnerabilities, and it has made defensive cybersecurity one of the model's headline capabilities.

Google also describes several safety systems being strengthened before broad availability.

These include:

• safeguards against cyber and CBRN misuse
• improved resistance to indirect prompt injection
• monitoring of the model's chain-of-thought and actions so execution can be stopped when necessary
• work to monitor internal activations for signs of misuse
• hardened, isolated environments for high-risk training and evaluations

There is an important exception.

Google says trusted cyber defenders and its own internal teams may receive Argon without cyber guardrails so they can use the model's full defensive cybersecurity capabilities.

That does not mean unrestricted public access.

The Fairwind Program is controlled access for vetted organizations, with governance and security requirements around how those capabilities are used.

Google's Fairwind documentation explains the broader trusted-defender model.

The hardened sandbox language also needs to stay in context.

Google is describing safeguards around its own training and evaluation environments, not advertising a user-facing sandbox feature inside Argon.

What happened to Gemini 3.5 Pro?

Gemini 3.5 Pro never became the public flagship Google originally appeared to be preparing.

In July, Google said it was testing 3.5 Pro with partners.

Later in September, Google DeepMind chief Koray Kavukcuoglu said the company had taken "a little bit of a step back" from the effort while focusing on Flash models and Gemini 4.

Subsequent reporting has gone further.

Bloomberg has reported that Google dropped the planned Gemini 3.5 Pro release, and Argon now effectively occupies the next-frontier slot that 3.5 Pro might once have filled.

Google itself has not, at publication time, issued a standalone announcement formally cancelling Gemini 3.5 Pro.

So the clean version is:

Google never shipped Gemini 3.5 Pro. Reporting says the planned release was dropped, while Google's own public comments had already confirmed that it stepped back from the model before focusing on Gemini 4.

For another recent Gemini change, see Choosely's Gemini 3.8 Live Avatar analysis.

What Google still hasn't said

The Argon announcement answers a lot of the questions this page was originally tracking.

It does not answer all of them.

As of October 1, Google still has not publicly provided:

• a general Argon API model ID
• a broad API availability date
• a general consumer availability date
• a published production input-context specification
• an end date for introductory pricing
• normal batch, Flex or Priority pricing
• a standard cache-storage price
• a broad-access model card surfaced through the normal developer documentation

That distinction matters because an announced model and a generally available product are not the same thing.

The next meaningful update to this page should come when one of those gaps closes.

Should you wait for Gemini 4 Argon?

For most people, no.

If your current Gemini, Claude or ChatGPT setup is doing the work you need, Argon's announcement is not a reason to freeze your workflow.

Use the model you can actually access today.

If you are a developer, keep your model selection configurable and preserve a small set of real tasks that you can rerun when Argon becomes available.

If you are already using Gemini, the Google Gemini profile on Choosely provides the current tool-level reference.

For a broader current-model decision, see Which AI Model Should You Use in 2026?.

The Choosely verdict

Gemini 4 Argon is now official.

That is the major change since this article was first published.

Google has moved from talking about an unreleased Gemini 4 training run to naming the model, publishing launch pricing, revealing a 1-million-token output limit and putting a broad benchmark package on the record.

The catch is access.

Argon is currently rolling out through a controlled early cohort rather than the public Gemini API or ordinary consumer channels.

Google's reported benchmark results are strong, but they are not a clean sweep and they are not a substitute for real-world testing once the model becomes accessible.

Gemini 4 has arrived as a product announcement. Its broad release date still hasn't.

That is the part worth watching next.

Frequently asked questions

Is Gemini 4 released?

Google has officially announced Gemini 4 Argon and begun a limited rollout to trusted cyber defenders through its Fairwind Program. It is not broadly available through the normal Gemini API or consumer products yet.

What is the Gemini 4 release date?

Google announced Gemini 4 Argon on September 30, 2026 US time. It has not announced a date for broad developer or consumer availability.

What is Gemini 4 Argon?

Gemini 4 Argon is Google's newly announced frontier Gemini 4 model. Google positions it for complex software engineering, enterprise knowledge work and defensive cybersecurity.

How much will Gemini 4 Argon cost?

Google says Argon will launch at an introductory price of $2 per million input tokens and $10 per million output tokens. Cached input is priced at 95% off the input-token rate. Google also states $4 input and $20 output per million after the introductory period, but has not announced when that period ends.

Does Gemini 4 Argon have a 1-million-token context window?

Google has announced a 1-million-token output limit. It has not published a general 1-million-token input or total context-window specification.

Can developers use Gemini 4 Argon now?

Not through the normal public Gemini API at publication time. Google says broader access will come later, beginning with paid API customers and Google AI Ultra subscribers.

Is Gemini 4 Argon better than GPT-6 Astra?

Google's benchmark table shows Argon ahead of Astra on several evaluations and behind it on others. Those are vendor-reported benchmark results. A broader real-world comparison should wait until Argon is publicly accessible and reproducibly testable.

What happened to Gemini 3.5 Pro?

Google never released Gemini 3.5 Pro. Bloomberg has reported that the planned release was dropped. Google's own earlier comments said it had stepped back from 3.5 Pro while focusing on Flash models and then Gemini 4.

Keep your Gemini stack current

Gemini 4 Argon has turned a future model into a real stack decision, but most users cannot make that decision yet.

Choosely Stack Intelligence helps keep the AI tools you rely on visible as models, pricing, access and capabilities change.

Review your AI stack →