Gemini 4 Argon review in short: this is Google’s new flagship model, built for long, complex work such as large codebases, legal and financial documents, cybersecurity and research. Since October 13, 2026 it has been available to Google AI Pro and Ultra subscribers in the Gemini app, and to developers on the paid Gemini API at an introductory $2 per million input tokens and $10 per million output tokens. For most people it’s worth it only if you already pay for Pro or Ultra and give Gemini heavy tasks. For everyday questions, the models you already have are enough.
Good to know: this is a first look based on Google’s official information, not a hands-on test. We didn’t pay for a subscription and didn’t test answer quality. We ran only one short API call to confirm the model responds. We checked every detail on October 13, 2026 on the Google blog, the Google DeepMind model page, the pricing page and the Gemini API docs.
At a glance
- What it is: Google’s flagship model for long tasks, code, professional work and cyber defense.
- Where: the Gemini app and AI Mode in Search for Pro and Ultra subscribers, Google Antigravity for Ultra, and the paid Gemini API tier.
- API price: $2 input and $10 output per million tokens through November 30, 2026, then $4 and $20.
- Free: no. There’s no access on the free app tier or the free API tier.
- Not tested: quality, speed and how much usage each plan gets.
We covered the launch itself in our news story on Gemini 4 Argon availability, and the original announcement in our report on Argon’s cyber-defender preview. This review answers a different question: does it change anything for you?
What is Gemini 4 Argon?
Argon is the first model in the Gemini 4 family. Google announced it on September 30, 2026, but at first opened it only to a group of cyber defenders through the Fairwind Program. On October 13, 2026 it opened to everyone. According to Google, it’s built for deep, multi-step work: refactoring large codebases, analyzing legal and financial documents, finding security vulnerabilities and scientific research. Google says this is just the beginning, and more Gemini 4 models will follow.
In the app, Google says you should notice the style too: Argon answers more directly with less hedging, doesn’t repeat itself more than necessary and skips clichés. On longer tasks it shares its progress more often.
Who gets Gemini 4 Argon, and where?
| Who | Where it’s available | Note |
|---|---|---|
| Google AI Pro subscribers | Gemini app and AI Mode in Search | Plans page says it’s rolling out to Pro in select countries |
| Google AI Ultra subscribers | Gemini app, AI Mode, Google Antigravity and AI Studio | Ultra gets “higher access” than Pro |
| Developers | Paid Gemini API tier, AI Studio and Android Studio | Not available on the free API tier |
| Enterprises | Gemini for business, the API and Google Antigravity | Through the Gemini platform on Google Cloud |
| Researchers and cyber defenders | Free Antigravity for 6 months, or Fairwind | By application, subject to selection |
One detail worth knowing: the Gemini API pricing page says Argon isn’t available on the free tier or on the Google AI Pro plan. So Pro subscribers get it in the app and in Search, but not in AI Studio. To build with it, you need a paid API tier or an Ultra subscription.
The new Tasks mode is available, per Google’s help page, wherever Gemini is available except the European Economic Area, Nigeria, Switzerland and the United Kingdom. It also doesn’t work with work or school accounts.
How much does Gemini 4 Argon cost in the API?
| Item | Through Nov 30, 2026 | From Dec 1, 2026 |
|---|---|---|
| Input | $2 | $4 |
| Output (incl. thinking tokens) | $10 | $20 |
| Cached input | $0.10 | $0.20 |
The cached-input discount is 95%, and it applies to implicit caching only. Explicit caching isn’t supported for this model. Grounding with Google Search includes 5,000 free requests a month, shared across all Gemini 3 and newer models, then $14 per 1,000 requests.
What it means in practice: Argon is a relatively expensive model, and the discount ends in less than seven weeks. If you’re building a product on it, budget at the December price, not the launch price.
The 1M unified token limit and the continuation protocol
The standout technical change is a single 1,048,576-token limit shared across input, thinking and output. Most models have a large input window and a much lower output ceiling. Per the docs, a single Argon call generates up to 64K tokens. When an answer needs more, the server continues it in 64K hops, each picking up where the last one stopped, and combines them into one response. Google calls this the continuation protocol, and the docs cover it under long decode.
The docs describe two ways to use it: background execution on the Interactions API, where the server manages the hops for you, or client-side continuation on the GenerateContent API, where your code starts each hop. The input repeated in each hop gets the 95% caching discount. Google lists uses like long mathematical proofs, generating or refactoring a whole repository in one pass, and translating large legacy files.
What’s new in the Gemini app?
- Tasks toggle: in the chat window you can switch from chat to tasks. In that mode Gemini keeps working for you even when you’re away from your computer, across Google apps and more.
- More than 40 connected apps: Google names Instacart, OpenTable, Adobe and Shopify, among others.
- Code inside the app: Google Antigravity’s coding interface is now available directly in the Gemini macOS app.
If you still use Gems, note that they’re being replaced by Skills. See our guide to moving from Gems to Skills.
What Google’s benchmarks show
The table shows some of the rows on the Google DeepMind model page, using the names shown there. These are Google’s own numbers, not an independent test.
| Benchmark | Gemini 4 Argon | GPT-6 Astra | Claude Fable 5.1 | Claude Opus 5.5 |
|---|---|---|---|---|
| Vals Index | 68.9% | 63.1% | 65.8% | 67.0% |
| AutomationBench | 51.3% | 41.4% | 31.4% | 42.5% |
| Vals Finance Agent v2 | 65.4% | 53.5% | 58.9% | 58.6% |
| Harvey’s Legal Agent Benchmark | 19.6% | 5.4% | 6.7% | 3.8% |
| DeepSWE v1.1 | 77.9% | 74.1% | 67.4% | 74.2% |
| FrontierSWE v2 | 55.0% | 65.5% | 56.3% | 62.3% |
| Terminal-bench 4.0 | 57.4% | 58.2% | 57.9% | 66.4% |
| GraphWalks (256k to 1M, BFS) | 84.2% | 71.8% | 65.0% | 66.8% |
| OSWorld-2.0 | 69.2% | 72.6% | — | — |
| CWE-bench v1 | 68.0% | 68.0% | 58.0% | 67.0% |
Argon leads most rows, especially professional work such as Vals Index and Harvey’s Legal Agent Benchmark, and long context. But it doesn’t lead everywhere: GPT-6 Astra leads on FrontierSWE v2 and OSWorld-2.0, Claude Opus 5.5 leads on Terminal-bench 4.0, and CWE-bench v1 is a tie with GPT-6 Astra. If you’re picking a model for terminal coding, don’t assume Argon is automatically the right choice. For a wider look at tools, see our Cursor vs Claude Code comparison.
Cyber and safety: Fairwind and Google’s claims
The Fairwind Program gives organizations that defend critical infrastructure, such as governments, healthcare providers and telecom services, early access to Argon. According to Google, participants get the model without the usual cyber guardrails and are allowed only defensive and research work, such as malware analysis. Access can’t be sold or shared. Researchers and scientists can apply for the highest usage tier in Antigravity free for 6 months, subject to eligibility, capacity and selection.
On safety, Google says Argon is its most resilient model yet against indirect prompt injection, and that on the Gray Swan benchmark attack success dropped to 0.7%. It also says that in coding tests the model’s attempts to take shortcuts to finish a task dropped to 2%, and that a monitor reads its reasoning and actions and stops it when needed.
Gemini 4 Argon in one paragraph
Gemini 4 Argon is Google’s new flagship model for long, hard tasks, and it isn’t free: in the Gemini app it’s only for Pro and Ultra subscribers, and building with it requires the paid API, at a price that doubles after November 30, 2026. If you use Gemini for questions and emails, there’s no reason to upgrade for it. If you give it big codebases, long documents or tasks that run for hours, this is the model to try.
Who is Gemini 4 Argon for, and who should skip it?
- Yes: developers working on large codebases, legal, finance and research professionals who work with long documents, and security teams.
- Maybe: anyone already paying for Google AI Pro. You get it at no extra cost, so try it on one heavy task.
- No: people happy with the free tier, and anyone who only wants to prototype for free in AI Studio.
If you’re still deciding which chatbot to pay for at all, read our full Gemini review and our ChatGPT vs Claude vs Gemini buying guide.
What we still don’t know
- How good Argon’s answers are in everyday use. We haven’t tested it.
- How much Argon usage each plan includes. Google says “access” for Pro and “higher access” for Ultra, without numbers.
- When exactly the model will reach every Pro subscriber, since the rollout is gradual.
- Whether pricing will change after December, or whether cheaper Gemini 4 models will arrive.
New to a term? Look it up in our AI glossary.
Gemini 4 Argon FAQ
What is Gemini 4 Argon?
Google’s new flagship model, built for long, complex work: code, legal and finance tasks, cybersecurity defense and research. It was announced on September 30, 2026 for a group of cyber defenders and opened to everyone on October 13, 2026.
Can I use Gemini 4 Argon for free?
No. In the Gemini app it’s available to Google AI Pro and Ultra subscribers. In the API it isn’t available on the free tier, only on the paid tier.
How much does Gemini 4 Argon cost in the API?
Through November 30, 2026: $2 per million input tokens and $10 per million output tokens. From December 1, 2026: $4 and $20. Cached input is 95% cheaper.
What is the 1M unified token limit?
In the API, Argon has a single window of 1,048,576 tokens shared across input, thinking and output. One call generates up to 64K tokens, and longer answers are produced in continuation hops, so the output can be very long.
Can Google AI Pro subscribers use Argon in AI Studio?
No. The API pricing page says Argon isn’t available on the free tier or on the Google AI Pro plan. Pro subscribers get it in the Gemini app and AI Mode in Search, while Ultra subscribers can also try it in AI Studio and Google Antigravity.
What is the Fairwind Program?
A Google DeepMind program that gives approved cyber defenders, such as governments, healthcare providers and telecom services, early access to Argon. According to Google, participants get it without the usual cyber guardrails, for defensive work only.
Gemini 4 Argon review: the bottom line
Argon is a big step for Google on long tasks and professional work, at least by the numbers Google itself publishes. The decision is fairly simple: if you already pay for Pro or Ultra, try it on your heaviest task. If you’re a developer, budget at the December price. And if you’re on the free tier, there’s nothing here for you yet.
Sources
- Google Blog: “Gemini 4 Argon: our next era of frontier intelligence,” September 30, 2026
- Google Blog: “Gemini 4 Argon: Built to solve your hardest problems,” October 13, 2026
- Google Blog: “Gemini 4 Argon is now available for developers,” October 13, 2026
- Google Blog: “Gemini 4 Argon is now available in the Gemini app,” October 13, 2026
- Google DeepMind: Gemini page, performance table and Fairwind Program page, checked October 13, 2026
- Gemini API: pricing page, models list and long decode docs, checked October 13, 2026
- Google One: Google AI plans page, checked October 13, 2026
- Gemini Apps Help: Tasks help page, checked October 13, 2026
