Gemini 4 Argon Is Live, and Its API Price Doubles on December 1

Google made Gemini 4 Argon, its new flagship model, widely available on October 13, 2026: in the Gemini app for Google AI Pro and Ultra subscribers, and in the paid tier of the Gemini API. Developers pay an introductory $2 per million input tokens and $10 per million output tokens, and Google's pricing page says both rates double on December 1, 2026.

Two weeks from announcement to your app

Google announced Argon on September 30 but held it back at first. The opening group was a set of trusted cyber defenders in its Fairwind Program, because the model can find and patch serious software flaws on its own. Two weeks later it opened up.

What Google says it can do

The announcement aims Argon at three kinds of work: real software engineering, office work such as legal and finance, and cyber defense. The most unusual change is length. Argon can now write up to about 1 million tokens in a single answer, up from 64,000, so it can think through a large problem and deliver the whole result in one go, such as a full codebase migration or a very long report.

These are Google's own numbers, and every lab picks tests where it shines. The independent picture is still forming: in our AI model leaderboard snapshot, Argon has the top score on LMArena's text board, in a statistical tie with Claude Opus 5.5 and Meta's Muse Wonder, and it ranks third of the 24 language models we track on Humanity's Last Exam, as measured by Artificial Analysis. It has fewer than 5,000 arena votes so far, so treat early ranks as provisional.

What changes in the Gemini app

For subscribers, the Gemini app update matters more than the benchmarks. Google says Argon answers more directly, hedges less, repeats itself less and skips clichés. Three features came with it:

The Shopify connection is the one to watch if you sell online. Google says the connected apps help with everything from booking a table to building a business, but it has not published a full list of what Gemini can change inside each app.

Its release notes also say most connected apps work in English in the US, with some in other English-speaking regions, Brazil and India. Start with read-only questions before you let it act.

If you build on the API: do the math before December

Argon is billed per token like every model API. The jump on December 1 matters most if you plan to run it inside automations, where volume adds up quietly. Here is how its output price compares with the other flagship models today:

Cached input also doubles, from $0.10 to $0.20 per million tokens, and Google's pricing page states plainly that Argon is not available on the free tier. If you only want to experiment, a Google AI Pro subscription in the app is the simpler route.

How to try it this week

In the Gemini app, flip the toggle from chat to tasks for something that spans several steps. A good first test uses your own files, because that is where long-context models earn their keep:

Go through the files in my Google Drive folder [FOLDER NAME] and build one summary document.
For each file, give me the date, the main decision or number, and anything that conflicts with another file.
Put a short list of open questions at the top. Show me the document before you share it with anyone.

Check two things in the result: whether it caught conflicts between files, and whether every date and number matches the source. If you also code, our comparison of Claude Code and Gemini CLI explains how Google's developer tools stack up against Anthropic's.

What to watch next

• Free users. Google's launch post says the wider release starts with paying customers, and the October 13 update covers only Pro and Ultra subscribers. Watch for news about free accounts.
• Independent speed tests. Artificial Analysis had not published a speed measurement for Argon as of October 18, so real-world latency is still unknown.
• The price hike. December 1 is a fixed date on Google's pricing page. If you compare models for a project, compare them at the post-December price.