Gemini 4 Argon is here: who can use it, what it costs and how to start

What is Gemini 4 Argon?

Gemini 4 Argon is Google's most advanced AI model and the first of the Gemini 4 generation. Google DeepMind announced it on September 30, 2026, in a post by Koray Kavukcuoglu, Google's Chief AI Architect, describing it as the company's frontier model for real-world coding, enterprise knowledge work such as legal and finance, and cyber defense.

At launch, though, almost nobody could use it. According to Google's official announcement, Argon first went to a small group of vetted cyber defenders through the Fairwind Program, as part of a phased rollout driven by its cybersecurity capabilities. The news this week is that this phase is over: since October 13 the model is open to paying subscribers and developers.

A frontier model is simply a model operating at the highest capability level available at a given time. For context on this month's model race, we also covered the launch of Meta Muse Wonder, Meta's first frontier model.

Who can use Gemini 4 Argon, and where

On October 13, Google opened Argon on several fronts at once:

• Gemini app: according to Google's blog, it is available to Google AI Pro and Google AI Ultra subscribers. As Thurrott reports, Ultra comes with higher access to the model.
• AI Mode in Google Search: per Search Engine Journal, Pro and Ultra subscribers can pick it by tapping the "+" icon in the "Ask anything" bar and choosing Argon from the model menu. The rollout is gradual, so the option may take a few days to appear.
• Developers: Argon is available on the paid tier of the Gemini API in Google AI Studio and Android Studio. Ultra subscribers can also try it in AI Studio and in Google Antigravity, Google's agent-first development platform.

Google's announcements do not list the countries where the model is live. The quickest way to check is to open the Gemini app on a Pro or Ultra plan and see whether Argon shows up in the model picker.

What changes in the Gemini app with Argon

Alongside the new model, Google updated the Gemini app. Here is what it announced:

• More direct answers: Google says Argon hedges less, repeats itself less and avoids clichés.
• Clearer long tasks: on long jobs the model shares its progress more often.
• Tasks mode: a toggle in the main window switches from chat to tasks, where Gemini can keep working while you are away from your computer.
• Coding: the Gemini app for macOS now includes Google Antigravity's agentic coding interface, so you can move between chat and code in one place.
• Connected apps: Gemini connects to more than 40 apps, including Instacart, OpenTable, Adobe and Shopify, for jobs like ordering groceries, booking a table or building an online store.

Gemini 4 Argon pricing and limits for developers

For developers, the model ID is gemini-4-argon. Google's documentation lists a unified 1 million token context window and long outputs of up to 1 million tokens; in its announcement Google says the output limit went up from 64,000 to 1 million tokens. A token is a chunk of text, on average a little less than one word.

These are the rates on the Gemini API pricing page, in US dollars per million tokens:

As a rough example, a request that sends 100,000 tokens and gets 20,000 back costs $0.20 + $0.20, about 40 cents through November 30 and about 80 cents from December 1. There is no free API tier for Argon.

How good is Gemini 4 Argon? The reported benchmarks

In its September 30 announcement Google cites several results. These are company-reported numbers, so treat them as such until independent tests catch up:

• DeepSWE v1.1 (software engineering): 77.9%, which Google calls a new state of the art.
• Zapier's AutomationBench (automation): first place at 51.3%.
• LVBench (long video understanding): 91.7%.
• CWE-bench v1 (vulnerability remediation): 68%, tied for first.

Google also describes internal uses: Argon-based agents are migrating C/C++ code to Rust at a scale of up to the 800,000-plus line Fuchsia Zircon kernel (work still under review before production), and they produced a libgav1 video decoder that runs 2.7 times faster than the existing Rust port. On safety, the company points to refusal training for dangerous cyber and CBRN requests, monitoring of the model's reasoning, and a leading result on the Gray Swan indirect prompt injection benchmark.

The long-video score matters for content creators: a model that can follow an entire video can help you summarize it, find key moments and draft captions or descriptions.

What it means for you: how to use Gemini 4 Argon today

If you already pay for Google AI Pro or Ultra, you can put it to the test today. Here are four practical exercises:

1. Head-to-head test. Take a task you do often (a content calendar, a client proposal, a video script) and give it to both Argon and the model you normally use. Judge the output on your own work, not on leaderboards.
2. A long job in Tasks mode. Try a prompt like: "Analyze these 10 articles about my industry, pull out the 5 recurring themes and give me 5 short-video ideas, each with a title, an opening hook and an outline." Let Tasks mode run and watch the progress updates.
3. Research in AI Mode. In Google Search, pick Argon from the "+" menu and ask a complex question about your market. Always check the sources it surfaces.
4. For developers. Use the lower rates through November 30 to run real tests in AI Studio and measure the cost per request before prices double in December.

A text model is only the first link in the chain: ideas still need to become videos, images and avatars. For the video side of the Gemini ecosystem, see our Gemini Omni guide, and The AI Creator System course walks you step by step through producing professional content with AI.