Google's Gemini 4 Argon arrives in the Gemini app and AI Mode

Google's Gemini 4 Argon is now live in the Gemini app and in AI Mode in Search, but only Google AI Ultra and Pro subscribers can reach it for now. The wider release caps an unusual rollout. Argon first went to members of Google's Fairwind Program earlier this month — governments and trusted partners that need models with the company's most advanced cybersecurity capabilities. Now the model is spreading to everyone else, and the pricing tells its own story about where Google thinks this model sits in the market.

Developers can access Argon today through the Gemini API and through Antigravity, Google's agentic software developer platform. Consumers get it through the Gemini app and AI Mode in Search, provided they pay for one of Google's subscription tiers. In the US, Google AI Pro costs $20 a month. The Ultra tier starts at $100. That makes Argon one of the more expensive models to reach general availability this cycle — and one of the cheaper ones to actually run.

What does Argon cost?

The API pricing is the number developers will notice first. Argon costs $2 per million input tokens and $10 per million output tokens. Those are introductory prices, and they expire on November 30. After that date, Argon doubles to $4 per million input tokens and $20 per million output tokens.

The doubling matters for anyone building on the model. A product that pencils out at introductory rates may not survive the permanent price. Google has effectively given developers a window — buy in cheap now, or budget for twice the cost later.

The subscription math works differently. A Google AI Pro subscriber in the US pays $20 a month and gets Argon inside the Gemini app and AI Mode in Search without per-token billing. Ultra subscribers pay at least $100. Google has not said whether Argon's availability will extend to free-tier users.

How does it compare to OpenAI?

The strongest external validation comes from Artificial Analysis, an AI benchmarking company. According to its figures, Gemini 4 Argon matches OpenAI GPT-6 Astra's score on the Artificial Analysis Intelligence Index — a composite across multiple AI benchmarks — at 60 percent of the cost per task at Argon's introductory prices.

That is a concrete claim, and it frames the competitive stakes. Matching a frontier rival on a composite benchmark while charging 60 percent of the per-task cost is the kind of arithmetic that moves enterprise procurement conversations. It is also, notably, tied to the introductory pricing. Once prices double on November 30, that cost advantage narrows unless performance or pricing elsewhere shifts too.

Google skipped a generation to get here. The company was supposed to launch a 3.5 Pro model, but it moved on and released Argon instead. At launch, Google said the model can handle complex tasks involving:

• finance,
• software engineering,
• coding,
• creative writing,
• and cybersecurity defense.

That list reads like a bid for enterprise and developer workloads rather than casual chat. The cybersecurity angle explains the early Fairwind access. Governments and trusted partners got Argon first precisely because Google built it with the company's most advanced cybersecurity capabilities.

What changes inside the Gemini app?

Josh Woodward, VP of Product for Gemini and Google Labs, described the practical improvements in a launch statement. In the Gemini app, he said, Argon answers more quickly and with less hedging, repetitions and clichés. Anyone who has waited through a model's throat-clearing before it gets to an answer will understand why Google is leading with that.

The bigger change is how Argon handles long tasks. It can return responses while it's still thinking when dealing with longer tasks — tasks that would usually mean waiting a bit, or a long while depending on the AI being used, for results. That partial-output behavior shortens the perceived wait on exactly the kinds of complex jobs Google says Argon is built for.

Google is also reworking the app around it:

• It is making it easier to toggle between chat and tasks within the Gemini app.
• It is adding Antigravity's agentic coding interface inside Gemini's macOS app.

The Antigravity addition is the tell. Bringing an agentic coding interface into the consumer macOS app narrows the distance between Gemini the assistant and Gemini the development environment. Developers who start a task in chat can finish it in an agentic workflow without leaving the app.

Why does the rollout order matter?

The sequence — Fairwind first, then developers, then subscribers — is itself information. Google gave governments and trusted partners first access to a model it describes as having its most advanced cybersecurity capabilities. That is a security-first launch posture, and it aligns with growing government demand for AI models evaluated for defense use before general release.

Then came the API and Antigravity, Google's agentic developer platform, before the consumer app. Google built Argon's commercial case on developers and benchmark performance first. The consumer launch through AI Ultra and Pro subscriptions follows.

The stakes are straightforward. Google and OpenAI are trading benchmark parity and price cuts at the top of the model market, and Artificial Analysis's numbers put Argon at parity with GPT-6 Astra on its Intelligence Index at 60 percent of the per-task cost. For buyers, that turns the frontier-model decision into a procurement calculation as much as a capability judgment.

What happens after November 30?

The introductory pricing deadline is the date to watch. At $2 per million input tokens and $10 per million output tokens, Argon undercuts rivals on cost per task at comparable benchmark performance. At double those rates, the value proposition depends on whether Argon's real-world performance holds up where composite scores don't fully measure it — long agentic tasks, cybersecurity work, and the code-and-writing jobs Google says it targets.

For now, Argon is live where Google said it would be: the Gemini API, Antigravity, the Gemini app, and AI Mode in Search, behind the $20 Pro and $100-and-up Ultra tiers. Woodward's pitch — faster answers, less hedging, responses that arrive while the model is still thinking — is aimed at the frustration users actually feel. Whether the doubled prices stick after November 30, or whether competition forces them back down, will say a lot about how the frontier-model pricing war settles.