Gemini 4 Argon Lands in the Gemini App: What Pro Subscribers Get

For two weeks, Gemini 4 Argon was mostly an announcement: at first, only selected security teams were allowed to use Google’s new frontier model. As of October 13, that is changing. Argon is now available to all Google AI Pro and Ultra subscribers in the Gemini app, and to paying developers through the API. For users, that means Google’s strongest model is no longer reserved for the most expensive plan. For developers, a clock starts today, because the low introductory price ends on November 30.

Key takeaways

• Since October 13, Gemini 4 Argon has been rolling out globally to Google AI Pro and Ultra subscribers in the Gemini app and in AI Mode in Google Search. Users without a subscription and those on the smaller AI Plus plan do not get it for now.
• New in the app is a Tasks mode, in which Gemini keeps working on assignments even when you are away from your computer.
• Google promises more direct answers with less hedging, fewer repetitions and fewer clichés.
• Through the API, Argon costs $2 per million input tokens and $10 per million output tokens until November 30. From December 1, all prices double; there is no free tier.
• 10,000 researchers can apply for six months of free access to the highest usage tier of Antigravity, Google’s agentic coding environment.

Who gets Argon now

The most important change concerns the cheaper of the two main plans. When Google unveiled the model on September 30, Ultra was the only consumer plan mentioned. Now Pro subscribers get access as well, in the Gemini app and in AI Mode in Google Search, the chat-style search view that has also been available in German since October 2025. In the United States, Pro costs $19.99 a month and Ultra $99.99. According to Google, Ultra offers “higher access” to Argon; Google has not yet published specific usage limits for either plan. Ultra subscribers can also use Argon in Antigravity, Google’s environment for coding agents.

Google describes the launch as global but is rolling the model out gradually. An Android Police writer with a Pro subscription did not see Argon in the app on day one. If you cannot find it, update the Gemini app and check the model picker. Free Gemini access and the entry-level AI Plus plan are left out for now. For security professionals, the Fairwind program continues in parallel, offering a version without the usual cyber safeguards.

What changes in the app

Google describes the everyday difference less in terms of benchmarks than in terms of answer style. Argon is meant to answer questions more directly, hedge less, repeat itself less often and avoid typical AI clichés. On longer tasks, it reports its progress more often and more clearly. That sounds unspectacular, but it addresses one of the most common complaints about chatbots: answers that spend three paragraphs framing the question before getting to the point.

The new Tasks mode is more noticeable. A toggle in the chat window switches from conversation to assignments that Gemini works through on its own, even after you leave your computer. It can draw on Google’s own apps along the way. For developers, Antigravity is now built directly into the Gemini app for macOS, so switching between chat and coding no longer requires a separate tool. How well Tasks actually handles longer workflows cannot be judged seriously on day one; independent hands-on tests of the app are still missing.

For developers: prices with an expiration date

In the Gemini API, the model carries the identifier gemini-4-argon. Until November 30, one million input tokens cost $2 and one million output tokens, including internal reasoning, cost $10. Cached input costs 10 cents, and batch processing halves the prices. From December 1, everything doubles to $4 and $20. That makes Argon more expensive than its predecessor Gemini 3.1 Pro, which costs $2 for input up to 200,000 tokens and $12 for output.

What a task ultimately costs, however, depends less on the list price than on the number of tokens consumed. In our launch analysis, Argon stood out for its high token consumption. The comparison platform Arena, by contrast, concludes from its own first tests that a task costs 70 percent less with Argon than with Claude Opus 5.5 and a third less than with GPT-6.1 Sol. Both can be true, depending on the type of task. In practical terms for teams: the seven weeks of introductory pricing are a good window to measure Argon against your own workflows. But budgets should assume the doubled price from December.

Context: strong, but not undisputed

In knowledge work such as law and finance, Argon tops several leaderboards, while competitors are partly ahead in command-line coding. What is new is that mixed signals are also coming from inside Google. According to a Bloomberg report, some engineers with access consider the model weaker on real coding tasks, especially user interface design, than its test scores suggest. Google calls it “inaccurate” to say that Gemini 4 is underperforming in coding. Google’s chief AI architect Koray Kavukcuoglu also indicated that the company wants to roll out an early version of the successor as soon as possible.

For Pro subscribers, the situation is simple: Argon costs nothing extra and can be compared with the previous model in everyday use right away, most revealingly with long documents and tasks that previously had to be broken into many separate steps. Occasional chat users will mainly notice the more concise tone. For developers, the clock is what counts: anyone planning to build Argon into a product should use the introductory price for measurement, not for budgeting.