Gemini 4 Argon lands in the Gemini app: what changes and who gets it

Google has started rolling out Gemini 4 Argon, the first model of its Gemini 4 family, in the Gemini app. Since 13 October 2026, it is available to Google AI Pro and Ultra subscribers in the app and in AI Mode in Search. Free users and Google AI Plus subscribers do not get it for now.

What Gemini 4 Argon changes in the app

Google announced Argon on 30 September, first for a small group of cybersecurity partners, then opened it to developers and to paying Gemini app users on 13 October. For everyday use, Google highlights a few changes:

• More direct answers, with less hedging, less repetition and fewer clichés, and better respect for the style you ask for.
• More frequent progress updates when it works on a long task.
• Much stronger coding, Google's main focus for this model.
• A one-million-token context window, shared between what you send and what it writes, enough for several long reports in a single conversation.

The app also gains a switch in the main chat window that moves from chat to tasks (the feature previously called Spark), which keep working on a job after you leave. Google's help page says tasks require Pro or Ultra and are not offered in the European Economic Area, the UK, Switzerland or Nigeria. Gemini can now also connect to more than 40 apps, such as Instacart, OpenTable, Adobe and Shopify, though most of them work only in English in the United States for now.

How Argon compares, according to Google

Google published a table comparing Argon with the latest top models from other AI labs. On the Vals Index, an independent score built from real professional tasks, Argon comes first:

Its clearest lead is on very long documents. In a test that asks a model to follow links scattered across 256,000 to 1 million tokens of text, Argon keeps much more of the information than its rivals:

It does not win everywhere. In Google's own table, Argon leads 14 of the 19 tests (one of them a tie), but it trails on some coding-agent and scientific-computing tests, such as FrontierSWE and Terminal-Bench. Several of Argon's scores were also calculated by Google itself.

Harder to trick

For an assistant that reads your emails and web pages on your phone, resistance to hidden instructions matters. In an independent prompt-injection test run by Gray Swan, attacks succeeded far less often against Argon than against Google's own Flash models:

What it costs

Argon is included only in the Google AI Pro and Google AI Ultra plans. Here are Google's monthly prices in the United States:

In France, Pro costs €21.99 a month. Usage limits are not fixed message counts: they refresh every five hours up to a weekly cap, and Pro gets four times the free allowance. Developers can also use Argon through the paid Gemini API, at an introductory $2 per million input tokens and $10 per million output tokens until 30 November 2026, then $4 and $20.

Should you pay for it?

Most people can keep the free app for everyday questions. Argon earns the Pro subscription if you work with long documents, code or detailed research. Google has not published how well it handles each language, so if you mostly write in French or Arabic, test it on your own work during the first month before you commit.

Will you upgrade to try Argon, or is the free Gemini app enough for what you do on your phone?