New Post

Gemini 4 Argon Went Global. Intro Pricing Ends November 30.

Gemini 4 Argon went globally available on October 13, 2026.

And the introductory API price already has an expiry date stamped on it. Developers and enterprises get the model through the API in AI Studio and the Gemini platform, Google AI Ultra subscribers get it in Google Antigravity. And both Ultra and Pro subscribers get it in the Gemini app and AI Mode in Search, per Google's announcement. The intro price is $2 per million input tokens and $10 per million output tokens. And it holds only through November 30, 2026, after which the listed rate steps up to $4 per million input and $20 per million output. Rollout: global on day one, but access is tiered by subscription, and the pricing is temporary. Those two facts should shape everything you do with this model in the next six weeks.

Where You Can Actually Get It

Google staged this rollout by tier, not by geography. The model is everywhere at once, but which surface you get depends on what you pay for.

| You are | Where Argon is available | |---|---| | Developer or enterprise on the paid API | Gemini API, AI Studio, Gemini platform, Android Studio | | Google AI Ultra subscriber | Antigravity, AI Studio, Gemini app, AI Mode in Search | | Google AI Pro subscriber | Gemini app, AI Mode in Search |

The developer path runs through the paid tier of the Gemini API in AI Studio and Android Studio. Google describes Antigravity as its agent-first development platform.

And it is an Ultra-only surface, which tells you where Google wants the serious build work to happen and which subscription it wants developers on.

Pro subscribers get the consumer surfaces: the Gemini app and AI Mode in Search, with the release notes confirming the global rollout to Pro and Ultra.

Read that table as a ladder.

Every tier gets a taste, and the interesting surfaces sit one rung up from where you probably are now. That is not a complaint, it is just the mechanics of the launch. And you should factor the subscription cost into any comparison you make against other vendors' API pricing.

The Intro Price Is a Deadline, Not a Discount

The pricing is the part of this launch that most coverage waves at and moves past.

Here is the whole picture, verbatim from Google's model documentation:

| Period | Input per 1M tokens | Output per 1M tokens | |---|---|---| | Through Nov 30, 2026 | $2 | $10 | | From Dec 1, 2026 | $4 | $20 | | Cached input | 95% off the input price | - |

Two things stand out.

First, the price doubles on both sides of the ledger on December 1, 2026. Second, cached input tokens carry a 95% discount, which is buried in the announcement but is arguably the more important number for anyone running repeated-context workflows.

Introductory pricing is a customer-acquisition instrument, not a gift. Google is buying six weeks of your usage data and your integration effort at half price, betting you will not rip Argon out of your stack when the rate doubles. The counter-move is simple: evaluate Argon at $4 and $20, not $2 and $10. If a workload only pencils out at the intro price, it never penciled out. I meter token spend on everything I ship precisely because intro rates made a bad workflow look profitable once and the renewal bill said otherwise.

The caching discount deserves its own test. Any pipeline that resends the same corpus, the same codebase, or the same agent scaffolding on every call is paying full freight for identical input. A 95% discount on cached input can flip a workflow from marginally viable to clearly worth keeping. And it works at both the intro and the permanent price.

What Argon Is Built For, and What to Feed It

Google calls this its latest, most capable model and says it can reason longer on hard problems for better results on complex coding, agentic, math, and science tasks. The Gemini app release notes frame it as a frontier model built to take complex tasks from start to finish, with named use cases like analyzing a complex financial report or creating a deck from scattered sources.

The technical specs that matter for planning: a unified 1M-token context window, decode of up to 1M tokens. And tunable thinking levels labeled low, medium, and high. That third one is the cost lever nobody prices in. A model designed to reason longer will spend more output tokens per answer. And output is the expensive side of this pricing table.

Run everything at high thinking and you will pay for depth you did not need; run everything at low and you bought a frontier model to do commodity work.

The context window changes what you can send, not just how much.

A 1M-token window means a whole codebase, a full contract set, or an entire reporting package can go in as one input, which is exactly where that 95% cached-input discount starts doing real work on the second call and every call after.

What I'd Run Before November 30

You have a fixed window between the October 13 rollout and the November 30 price change. And it is the cheapest this model will ever be to evaluate. Here is the benchmark plan I would run, in order:

- Pick two of your actual workloads, one coding or agentic, one document-heavy like the financial report analysis Google itself names. - Run each at low, medium, and high thinking, and log token counts and output quality for every run. The gap between medium and high is where you find out whether the extra reasoning earns its tokens. - Turn on caching for anything with fixed context and measure the difference on your input bill. - Project every result at $4 input and $20 output. That is the real price; the current one is a teaser rate.

The question this launch actually poses is not whether Argon is capable.

Google says it is the most capable model it ships. And the positioning around complex, start-to-finish work is consistent across every source.

The question is whether your specific workload earns the permanent price.

And the only honest answer comes from running your own jobs, not reading benchmark tables.

Run your two workloads this week while the meter is cheap, write down the token counts at each thinking level. And do the math at the December 1 price before you commit.

If you want a follow-up where I walk through that benchmark template line by line, say so in the comments and I will build it out.