The short answer: Gemini 4 Argon is Google’s new frontier AI model for coding, enterprise knowledge work and cybersecurity. Since 13 Oct 2026, developers on the paid tier of the Gemini API can use it in Google AI Studio and Android Studio, at an introductory price of $2 per million input tokens and $10 per million output tokens.
Key facts
- What: Google’s new frontier model, built for software engineering, legal and finance work, and cyber defence.
- Announced: 30 Sep 2026, first for a small group of trusted cyber defenders.
- Open to developers: 13 Oct 2026, on the paid Gemini API tier in Google AI Studio and Android Studio.
- Price: $2 in, $10 out per million tokens to start, rising to $4 and $20 after the introductory period.
- Free tier: none mentioned. Google AI Ultra subscribers can try it in AI Studio and Google Antigravity.
Some AI models launch with a keynote. Gemini 4 Argon launched with a bouncer.
When Google announced it on 30 Sep 2026, the guest list was short: a set of “trusted cyber defenders” in a programme called Fairwind. Everyone else got to read the blog post and refresh it. (Google)
On 13 Oct, the rope came down a little. Paying developers can now build with Argon through the Gemini API, in Google AI Studio and Android Studio. Google AI Ultra subscribers can play with it in AI Studio and in Google Antigravity, Google’s agent-first coding tool. (Google)
What is Gemini 4 Argon actually good at?
Google pitches Argon at three jobs: real-world software engineering, enterprise knowledge work like legal and finance, and cybersecurity defence. Google calls it a “frontier model for complex workflows”. In plain English, it is built for the long, boring, high-stakes tasks, not for writing your Instagram captions. (Google)
The benchmark numbers Google published, all company claims:
- DeepSWE v1.1 (real software engineering tasks): 77.9%, which Google calls a new state of the art.
- AutomationBench (business workflows across apps): ranked first at 51.3%.
- LVBench (long video understanding): 91.7%.
- CWE-bench v1 (fixing security flaws): tied for first at 68%.
The output limit is the quietly big one. Google says it jumped to 1 million tokens, up from 64,000. That is the difference between a model that writes you a function and one that can write back an entire migrated codebase in one go. (Google)
Why did Google keep it locked up first?
Because Argon is good at finding security holes. Google says it can autonomously find, validate and patch critical vulnerabilities, and that it shipped to trusted defenders and internal teams without its usual cyber guardrails. (Google)
One example from the announcement: security firm Wiz used Argon and found a critical vulnerability in healthcare software that earlier frontier models had missed. Great news if you are defending hospitals. Slightly terrifying if you are the hospital’s IT person who skipped last week’s patch. (Google)
The public version comes with the seatbelts back on. Google lists refusals for harmful cyber and CBRN requests, monitoring of the model’s reasoning and actions, and sandboxed testing before high-risk evaluations. (Google)
How much does Gemini 4 Argon cost?
The introductory price is $2 per million input tokens and $10 per million output tokens, with cached input 95% cheaper. After the introductory period, it doubles to $4 and $20. Google has not said when that switch happens, so read “introductory” the way you read “limited period offer” on a mattress website. (Google)
Fun coincidence: $2 in and $10 out is exactly what OpenAI charges for GPT-6 Sol on its API. The price war is not over. It has just agreed on a uniform. (OpenAI)
Our read (opinion)
The interesting part is not the model. It is the rollout. Release to defenders first, then to paying developers, then (eventually) to everyone. That staged approach is becoming the norm for powerful models, and frankly, it should.
It also tells you where the money is. Last week we wrote about GPT-6 turning ChatGPT answers into mini apps for a billion people. Google’s answer is to go straight for developers and enterprises first. Different playbooks, same destination: your company’s budget.
If you are not a developer, you will not touch Argon this month. But you will touch the apps built on it, sooner than you think.
Do this
If you write code or manage people who do, pick one painful, well-defined task this week (a library upgrade, a test suite, a security review of one service). Run it through Argon in AI Studio on a small budget, and compare the result and the bill against what you use today. Decide with your own numbers, not a benchmark chart.
FAQ
What is Gemini 4 Argon?
Gemini 4 Argon is Google’s frontier AI model for complex workflows, aimed at software engineering, enterprise knowledge work such as legal and finance, and cybersecurity defence. Google announced it on 30 Sep 2026.
Can I use Gemini 4 Argon for free?
Google has not mentioned a free tier. Developers use it through the paid tier of the Gemini API, and Google AI Ultra subscribers can try it in Google AI Studio and Google Antigravity.
How much does Gemini 4 Argon cost?
The introductory price is $2 per million input tokens and $10 per million output tokens, with cached input 95% off. After the introductory period it rises to $4 and $20.
When did Gemini 4 Argon become available to developers?
On 13 Oct 2026, through the paid Gemini API in Google AI Studio and Android Studio. Before that, it was limited to trusted cyber defenders and testers.
Is Gemini 4 Argon better than GPT-6?
It depends on the task. Google reports leading scores on benchmarks such as DeepSWE v1.1 (77.9%) and AutomationBench (51.3%), but these are company-published numbers. The honest test is running your own task on both.