---
title: "Google puts Gemini 4 Argon on sale: the frontier model now costs $19.99 a month"
description: "The first Gemini 4 model leaves its trusted-partner phase for Pro and Ultra subscribers and paying developers"
author: "Luis Goa"
published: 2026-10-13T16:00:01Z
modified: 2026-10-13T18:02:14Z
url: https://rews.cc/a/google-puts-gemini-4-argon-on-sale-the-frontier-model-now-co-400db7
language: en
tags: ["google", "gemini", "ai", "openai", "muse", "tech"]
publisher: "Rews (https://rews.cc)"
---

# Google puts Gemini 4 Argon on sale: the frontier model now costs $19.99 a month

*The first Gemini 4 model leaves its trusted-partner phase for Pro and Ultra subscribers and paying developers*

By Luis Goa · October 13, 2026 · https://rews.cc/a/google-puts-gemini-4-argon-on-sale-the-frontier-model-now-co-400db7

## In brief

- Gemini 4 Argon is rolling out to Pro and Ultra Gemini subscribers ($19.99 per month minimum) and paying developers
- Argon was unveiled last month but restricted to trusted partners for cybersecurity testing
- Google says internal use raised engineers’ AI-generated code production by 35%, without published methodology
- Google says it monitors Argon’s chain of thought so operators can intervene when behavior looks off
- The launch follows OpenAI’s GPT-6 rollout and Meta’s Muse Wonder announcement a day earlier

Access to Google’s newest AI model now starts at $19.99 per month. Gemini 4 Argon is rolling out to Pro and Ultra subscribers of the Gemini app — who will also see it in AI Mode in Google search — and to developers, who pay separately, [Hugh Langley reported for Business Insider](https://www.businessinsider.com/google-gemini-4-argon-ai-model-public-release-cost-2026-10). Free-tier users get nothing.

Google is positioning Argon as its workhorse for coding, cybersecurity and knowledge work in finance and law. Tulsee Doshi, Gemini product lead at Google DeepMind, called it a “powerhouse” and the first of a family of Gemini 4 models to roll out over time. When [Google announced the model last month](https://rews.cc/a/google-announces-its-first-gemini-4-model-with-benchmark-win-fe3d20), only trusted partners could use it, deliberately, to probe for cyber vulnerabilities first. The release matters more than most because Google’s previously promised frontier model, Gemini 3.5 Pro, never shipped.

## A 35% claim and a cornhole-playing robot

Google employees have tested Gemini 4 internally for several weeks, and the company says the model helped engineers increase production of AI-generated code by 35% while improving products such as Chrome. That 35% is Google’s own measurement of its own engineers, published without methodology — no baseline, no denominator, no definition of “production” — so treat it as a marketing figure until the underlying data appears. Doshi credits the internal testing program itself: “I actually think it’s been one of the biggest game changers for us,” she told Business Insider, saying daily use by Googlers surfaced the model’s strengths and weaknesses.

![A blue robotic hand holds a long accordion-folded paper strip emerging from an orange stapler against a yellow background.](https://rews.cc/img/bacb66b83a79ed20aa995f01f99c9b735f52f820.webp)

The robotics claim is more concrete in mechanism, if not in numbers: Doshi said Argon’s multimodal and reasoning improvements sped robotics work inside Google, and the company trained a robot to play cornhole by having the model watch a human player and compute joint movements and spatial coordinates.

On safety, Google says it tracks the model’s chain of thought to detect misbehavior, which in principle allows intervention mid-run. “You can say, ‘Okay, the model is actually veering off in a way that’s concerning. Let’s stop here,’” Doshi said. “I think that gives us a much more robust set of protections in launching this model.” Monitoring a chain of thought only helps if the reasoning the model writes down reflects what it is actually doing, and Google has published no evaluation showing that here.

Doshi confirmed the model rolling out is the same one whose benchmarks Google published last month — not whatever employees are testing now. Better versions already exist internally; one internal build [known to staff as “Carbon” was described as feeling like “Opus 5.5” for coding](https://rews.cc/a/google-staff-test-gemini-4-version-called-carbon-ahead-of-ar-0924f9), with no commitment it would ever ship. “People should expect to see continued improvements coming from us wholesale,” Doshi said.

The competitive clock is visible. Meta said Monday that its Muse Wonder class of models reaches the frontier and powers its Muse personal agent; OpenAI began rolling out GPT-6 last month. Google’s benchmarks for Argon remain self-reported, but the release changes who can check them: as of today, anyone with $19.99 can run the test.