Gemini 4 Argon: What Google Says It Can Do and Who Can Use It

A friendly robot with a glowing crystal on its chest works at a desk full of books, a tablet and a laptop, with floating screens showing an image, text, a chart and a video, while a person watches with an impressed smile

Written by

in

· Published

On October 13, 2026, Google announced Gemini 4 Argon, the newest model of its Gemini family. Google presents it as a model for hard, multi-step work: coding, finance, legal, cybersecurity and science. I have not used Argon myself, so this is not a review. It is a summary of what Google says in its announcement, in my own words, and a list of what I would check before believing it. I read the announcement with Claude’s help.

The short version

Google says Argon is built for deep, complex work rather than quick answers, that it can handle very long tasks, and that it writes more directly than earlier Gemini models. It is available to developers through the API, and to people through the Gemini app if they pay for a Google AI Pro or Ultra plan. The announcement does not mention a free option.

Who can use it, according to Google

You are…Where Google says you can use Argon
A regular userThe Gemini app and AI Mode in Search, with a Google AI Pro or Google AI Ultra subscription.
A developerThe Gemini API in Google AI Studio and Android Studio (paid tier). With a Google AI Ultra subscription, also Google Antigravity.
A companyGemini for business, the API and Google Antigravity.
A scientistGoogle offers 10,000 scientists the highest usage tier of Antigravity free for 6 months. Applications are open, with eligibility and selection.
A cybersecurity defenderApplications to the Fairwind Program, which gives access without the usual cyber safeguards.

The main claims

  • Price for developers: an introductory price of $2 per million input tokens and $10 per million output tokens, with cached input tokens 95% cheaper. “Introductory” means it may change.
  • Long tasks: Google mentions a 1 million token limit that covers input and output together in the API, and a new “continuation protocol” for very long jobs.
  • Benchmarks: Google says Argon ranks first on the Vals Index, which measures performance in areas such as finance, coding, legal and tax work. That is Google’s own statement about a ranking I have not checked.
  • Writing: Google says Argon answers more directly with less hedging, does not repeat itself more than necessary, and avoids clichés.
  • Safety: Google says it is better at refusing harmful requests and resists hidden instructions in web pages and emails more than earlier models. Its own test shows attacks succeeding 0.7% of the time across 15 adaptive attempts.

Examples Google gives

The announcement is mostly examples from Google’s own teams and early partners. Two that give an idea of the scale:

  • Google says its Chrome team used Argon to separate part of a large legacy folder of browser code, finishing in weeks what they expected to take months.
  • Google says its Agent Development Kit team gave Argon a goal overnight and got changes that cut some session preparation time by 40% and sped up some imports by 21%.

These are results reported by the company that makes the model, so they show what is possible, not what you will get with your own work.

What I find interesting for everyday use

Most of the announcement is about developers and companies, but one claim is useful for anyone who writes with AI: Google says Argon repeats itself less and avoids clichés. That connects with something I tested with ChatGPT, where I found that the same answers kept coming back. You can read it in why ChatGPT repeats itself. If Google is right, it would be a visible difference.

What I want to test

  • Does Argon really hedge less? I would give it the same question as ChatGPT and Claude and compare how direct the answers are.
  • Does it repeat itself less across several answers to the same prompt, as in my ChatGPT test?
  • Does it respect word limits better? I tested that with ChatGPT in does ChatGPT respect word limits.
  • Is it worth paying for a Pro or Ultra plan only for this model?

If you are also deciding between Claude models, I summarized Anthropic’s three new ones in the Claude Opus 5.5, Sonnet 5.5 and Haiku 5.5 guide.

What to keep in mind

  • Everything above comes from Google’s announcement. I have not tested Argon and I have not verified the benchmarks or the examples.
  • Plans, prices and availability change. The introductory price in particular may not last.
  • If something here no longer matches Google’s page, trust the page.

Source

About the author
I’m Christian Serrano, and I run AIVERSUSIA. I sell products online and use AI tools in that work every day. I test tools with the same prompts and publish what happened, including mistakes and limits. Claude helps me write articles and run tests, and I review and approve everything before it is published. More about me · Editorial policy