Gemini 4 Argon is Google’s newest and most capable AI model, announced on 30 September 2026. It is built for long coding tasks, office-style knowledge work and cybersecurity. Right now, most people cannot use it. Google is giving it first to a small group of security partners, then to paid API customers and Google AI Ultra subscribers. Google’s official announcement confirms this rollout order.
This post explains what Gemini 4 Argon does, what it costs, who gets it first, and what you can use today while you wait.
Gemini 4 Argon at a glance
| Detail | Gemini 4 Argon |
|---|---|
| Announced | 30 September 2026 |
| Made by | Google DeepMind |
| Main uses | Long coding tasks, enterprise knowledge work, cyber defence |
| Maximum output | Up to 1 million tokens per response (previous limit: 64,000) |
| Available now to | Selected security partners in Google’s Fairwind Program |
| Next in line | Paid Gemini API customers and Google AI Ultra subscribers |
| API price (introductory) | $2 per million input tokens, $10 per million output tokens |
| API price (later) | $4 per million input tokens, $20 per million output tokens |
| Cached input | 95% off the input price |
| Free Gemini app | No date announced |
What Gemini 4 Argon is good at
Google describes Argon as a model that can keep reasoning through long, multi-step jobs without losing track. In practice, that points to three kinds of work:
- Coding over many hours: fixing bugs across a large codebase, moving code from one language to another, and running long agent tasks. Google says it used Argon internally to help move C and C++ code to Rust.
- Knowledge work: legal research, finance analysis and reading long documents, charts and videos.
- Cyber defence: finding, checking and patching security bugs in software. This is why security teams get it first.
The 1 million token output limit is the headline number. One token is roughly three quarters of an English word, so a single answer could, in theory, run to hundreds of thousands of words or a very large code file. Most everyday chats will never come close to that.
Gemini 4 Argon benchmarks
These scores come from Google and were reported by VentureBeat. They have not yet been checked independently, so treat them as the company’s own claims.
| Benchmark (what it tests) | Gemini 4 Argon | GPT-6 Astra | Claude Opus 5.5 |
|---|---|---|---|
| DeepSWE v1.1 (software engineering) | 77.9% | 74.1% | 74.2% |
| AutomationBench (work automation) | 51.3% | 41.4% | 42.5% |
| Vals Finance Agent v2 (finance tasks) | 65.4% | 53.5% | 58.6% |
| LVBench (long video understanding) | 91.7% | 87.5% | 83.7% |
| Terminal-bench 4.0 (command line tasks) | 57.4% | Not listed | 66.4% |
| FrontierSWE v2 (hard coding tasks) | 55.0% | 65.5% | Not listed |
So Argon is not ahead everywhere. It leads on several agent, finance and video tests, but scores lower than rivals on some coding and terminal tests in Google’s own table. VentureBeat counted Argon leading or tying in 13 of 18 published benchmarks.
Gemini 4 Argon pricing
For developers, Argon is priced per token through the Gemini API:
| Price type | Introductory price | Price after the offer |
|---|---|---|
| Input | $2 per 1M tokens | $4 per 1M tokens |
| Output | $10 per 1M tokens | $20 per 1M tokens |
| Cached input | 95% off input | 95% off input |
Google has not said how long the introductory price will last. If you plan a project around it, budget for the higher price.
For regular users, the route is a Google AI Ultra subscription. Since May 2026, Ultra starts at $100 a month in the US, with a $200 tier that has higher limits. Prices differ by country. Google has not confirmed which Ultra tier gets Argon first or in which countries.
How to get access to Gemini 4 Argon
There is no sign-up button for the public yet. Here is what you can do now, depending on who you are.
Google has not published exact access steps yet, so the steps below are based on how earlier Gemini models were rolled out. Menu names may differ when Argon arrives.
If you are a developer
- Sign in to Google AI Studio with your Google account and make sure you have a billing-enabled (paid) API project. Free-tier keys are not mentioned in the launch plan.
- Open the model picker and look for a model name containing “Gemini 4 Argon”. It will not appear until Google opens access to your account.
- When it appears, run a small test prompt first and check your usage page, because the output limit is very large and long answers cost more.
If you use the Gemini app
- Open the Gemini app or gemini.google.com.
- Tap the model selector at the top of the chat.
- If you have Google AI Ultra and Argon has reached your account, it will show in that list. If it is not there, it has not reached you yet.
If you are on a free account or Google AI Pro, Google has not given any date. Keep using the current Gemini models in the app; they are still updated regularly.
Troubleshooting
- “Model not found” error in the API: your project does not have access yet, or the model name is wrong. Copy the exact name from the model list instead of typing it.
- Argon is not in the Gemini app model list: access is rolling out in stages. Signing out and back in will not speed it up.
- A website or app offers “free Gemini 4 Argon”: be careful. As of 1 October 2026, Google only offers it through its own Fairwind Program, API and AI Ultra. Third-party sites making that claim may be using a different model or collecting your login.
- API bill is higher than expected: set a maximum output token limit in your request. Without it, a long agent run can produce a very long answer.
Common mistakes
- Assuming Argon is in the free Gemini app. It is not, and no date has been given.
- Reading the benchmark table as proof it is best at everything. On some coding tests, rivals scored higher.
- Paying for AI Ultra only for Argon before Google confirms access for your country and plan.
- Building on the introductory price. The listed later price is double.
What this means for you
Example: you run a small software team and spend hours each week on bug fixing and code migrations. Argon is worth testing as soon as your paid API project gets access, starting with one well-defined task and a strict output token limit. Compare its results and cost against the model you use today.
If you mainly use Gemini to write emails, summarise documents or plan trips, nothing changes for you yet. The current Gemini models already handle that well. If you are in India and use Jio, see our post on getting Gemini Pro free with Jio. If you write in Word, our guide to the ChatGPT for Word add-in covers a similar tool you can use today.
Frequently asked questions
Is Gemini 4 Argon out?
It was announced on 30 September 2026, but only selected security partners can use it so far. Paid API customers and Google AI Ultra subscribers are next. There is no date for free users.
Is Gemini 4 Argon free?
No. At launch it is only for Fairwind Program partners, paid API users and Google AI Ultra subscribers. API pricing starts at $2 per million input tokens and $10 per million output tokens.
Is Gemini 4 better than ChatGPT?
On Google’s published benchmarks, Argon beats GPT-6 Astra on many agent, finance and video tests, but scores lower on some coding and terminal tests. Independent tests are still limited.
Is the Gemini API still free?
Google has not announced a free tier for Gemini 4 Argon. The launch plan names paid API customers only.
What is the Fairwind Program?
It is Google’s early access program for trusted cybersecurity defenders. They get Argon first to find and fix software vulnerabilities.
Final note: if you want Argon, the realistic path today is a paid API project or a Google AI Ultra plan, and even then you have to wait for the rollout. Bookmark Google’s announcement page, since that is where wider availability will be confirmed.
New to Gemini? Our Gemini AI assistance guide explains what the assistant can do and how to get started for free.

