Gemini 4 Argon Is Out: Which Google Plans Get It and What It Costs

Google released Gemini 4 Argon to paying users on 13 October 2026, two weeks after announcing it as its new frontier model. It's the first model in the Gemini 4 family, and on Artificial Analysis's independent index it ties OpenAI's GPT-6 Astra.

Where you can use it depends on which Google plan you have. In the Gemini app it needs Google AI Pro at $19.99 a month or AI Ultra, and the free plan and the $4.99 AI Plus plan don't get it. AI Pro unlocks it in the app and in Google Search, but not in Google AI Studio or the Antigravity coding tool. And the API price Google launched with is introductory. Its own pricing page doubles it on 1 December.

I read Google's three launch posts, its US plan page, the Gemini help centre's articles on models and limits, the Gemini API pricing page and Artificial Analysis's model pages on 13 October 2026. Everything below comes from those.

The Short Answer

Where	What you need	Price
Gemini app	Google AI Pro or AI Ultra	$19.99 a month. Ultra is $99.99 or $199.99
AI Mode in Google Search	Google AI Pro or AI Ultra	The same plans
Google AI Studio	A paid Gemini API key, or AI Ultra	Per token, or Ultra
Google Antigravity	AI Ultra	From $99.99 a month
Gemini API	The paid tier	$2 input and $10 output per million tokens until 30 November, then $4 and $20
izzedo chat	Any paid plan	From $6 a month

There's no free way in for most people. Google's plan page lists 3.5 Flash-Lite as the free plan's model and 3.6 Flash as AI Plus's, and its API pricing page marks Argon "Not available" on the free tier. The two exceptions are application programmes, one for scientists and one for security teams, and I've put them at the end of the AI Pro section.

What Argon Is

Google pitches Argon at long, multi-step work, and its launch posts keep coming back to three kinds of it: software engineering, professional work such as legal and finance, and cyber defence. On the API it can now write up to a million tokens in one response. Google calls this a "1M unified input-output token limit", so the prompt and the answer share that million. The previous output limit was 64,000 tokens.

The Gemini app's announcement says more about everyday use. Google says Argon "answers questions more directly with less hedging, doesn't repeat itself more than necessary, and skips clichés", and that on longer tasks it reports its progress more often. That's Google's description, and I haven't measured it. Hedging and repetition are the two habits I'd most like gone from chat models, though, so it's the first thing I'd test on my own prompts.

Google's benchmark claims are strong. It reports 77.9% on DeepSWE v1.1, a test of long software engineering tasks, first place on the Vals Index, which weights finance, coding, legal and tax work by their share of US GDP, and first place on Zapier's AutomationBench at 51.3%. Those are Google's own results. The independent scores are further down.

Who Gets It in the Gemini App

Google's help article on Gemini's models sets out which plans see Argon in the picker:

Plan	Price	Argon in chat	Argon in Tasks
No plan	$0	No	No
Google AI Plus	$4.99	No	No
Google AI Pro	$19.99	Yes	Yes
Google AI Ultra	$99.99 or $199.99	Yes	Yes

Tasks is Gemini's mode for handing it a whole job, and Google says it "can work for you even when you're not at your computer". On a work or school account, Google says the models you get depend on your organisation's Workspace licence.

Google publishes no Argon allowance. Gemini meters compute rather than counting prompts. Your limit refreshes every five hours until you hit a weekly cap, AI Pro gets four times the free plan's allowance, and Ultra gets five or twenty times AI Pro's. The models article adds that "more advanced models and higher thinking levels require more of your usage", so Argon on High uses more of it than any other choice in the picker. When a paid plan runs out, the conversation carries on with Flash-Lite until the limit refreshes.

One setting is worth changing before you judge it. Every model in Gemini's picker takes a thinking level of Low, Medium or High, and Google says it defaults to Low if you don't pick one. Artificial Analysis tested Argon on High, so a hard problem on the default setting isn't a fair test of the model.

The AI Pro Catch: Not in AI Studio or Antigravity

AI Pro is the cheapest Google plan with Argon. It covers the Gemini app and AI Mode in Search, and it stops short of Google's developer tools. The Gemini API pricing page puts it plainly: "Gemini 4 Argon is not available on the Free tier or on the Google AI Pro plan." To use Argon in Google AI Studio, an AI Pro subscriber needs a paid API key, where every token is billed, or Ultra.

Antigravity is murkier. The AI Pro card on Google's plan page lists "entry rate limits to agent model in Google Antigravity", but the launch post names only Ultra subscribers as getting Argon there. If coding with Argon in Antigravity is the reason you'd upgrade, the plan Google names is Ultra, at $99.99 a month.

The two free routes both need an application. Google is offering 10,000 scientists Antigravity's highest usage tier free for six months, "subject to eligibility, capacity, and selection", with applications open now. And security teams can apply to the Fairwind Program, which gives trusted cyber defenders Argon without its cyber guardrails. Neither is meant for everyday users.

What It Costs on the API, Before and After 1 December

Gemini API paid tier, per million tokens	Until 30 November 2026	From 1 December 2026
Input	$2.00	$4.00
Output, thinking tokens included	$10.00	$20.00
Cached input	$0.10	$0.20

A few details sit under that table on Google's page. There's no free tier for Argon at all. Caching is implicit only, so Google applies the discount when a request repeats the start of an earlier one, and you can't set up a cache yourself. Grounding answers in Google Search comes with 5,000 free requests a month, shared across every Gemini 3 and newer model, and then costs $14 per 1,000. And on the paid tier, Google says your content isn't used to improve its products.

From 1 December, Argon's rates match the $4 and $20 that Artificial Analysis lists for Claude Opus 5.5. Google calls the launch price "introductory".

Nothing Google has announced changes the price for Gemini app subscribers. The $19.99 and $99.99 are plan prices, and the 1 December date is on the API pricing page, not the plan page.

How Argon Scores Next to the Other Frontier Models

For an outside view, Artificial Analysis puts every major model through the same ten evaluations and reports a score, the bill for the run and the number of tokens written. These figures are from each model's own page, read on 13 October:

Model, at the setting tested	Intelligence Index	Cost per index task	Output tokens across the index
Claude Opus 5.5, max	58	$5.98	260M
Gemini 4 Argon, high	53	$1.99	110M
GPT-6 Astra, max	53	$3.26	60M
Grok 4.7, xhigh	46	$3.74	240M

Argon ties Astra for about 61% of the cost. Astra is the more concise model, writing 60 million tokens to Argon's 110 million, but each of its tokens costs five times as much, $10 and $50 per million against $2 and $10.

Opus 5.5 still scores highest of the four. It costs three times as much per task as Argon and writes more than twice as many tokens.

The 1 December rise changes the comparison. Google's schedule doubles every Argon rate, so on the same test its cost per task would roughly double too, to about $4. That's my arithmetic, not Artificial Analysis's. If OpenAI's prices stay where they are, it would make Argon more expensive per task than Astra for the same score, though still cheaper than Opus 5.5.

One index doesn't settle which model is best for your work. Artificial Analysis's leans toward agentic tasks, research and reasoning, and the cost figures only matter if you pay by the token. In the app, what you pay is the plan. Which of the three big chatbots to use for which job, and what each $20 plan runs, is in Claude vs ChatGPT vs Gemini.

Should You Upgrade for Argon?

• You already pay for AI Pro or Ultra. It's in your picker now. Use it, on High, for the long jobs Google built it for, like the one in its own demo, where Argon "takes an unstructured folder full of project context and formats it into a clean proposal deck for a client". Keep Flash for quick questions, because Argon draws on the same allowance faster.
• You're on the free plan or AI Plus. $19.99 buys Argon and the rest of AI Pro, including Gemini inside Docs, Sheets and Slides, a 1 million token context window in the app and 5 TB of storage. If you work in Google's apps all day, that's a fair price. Grok vs Gemini goes through what each rung of Google's ladder buys.
• You build on the API. You'll need a paid key either way. Price your product on the December rates, not the launch ones.
• You mainly want to see how it compares with ChatGPT or Claude. You don't need a Google plan to do that.

Gemini's picker has changed three times since May. How Flash, Pro and the thinking levels fit together is in Gemini Thinking vs Pro, and the head-to-head with OpenAI is in Gemini vs ChatGPT.

Using Gemini 4 Argon in izzedo

izzedo chat has Gemini 4 Argon in its model picker on every paid plan, starting with Hobby at $6 a month. It sits in the same thread as GPT, Claude, Grok, DeepSeek and the rest, so you can draft with one model and hand the conversation to Argon to check it, or the other way round, and whichever model answers sees the whole thread. Argon takes the same three reasoning levels here, low, medium and high.

Argon uses up a plan's allowance faster than lighter models like Gemini Flash, and the pricing page shows how far each plan goes. The free plan doesn't include Argon, but it does include Gemini 3.8 Flash, the Flash that Google's plan page keeps for AI Pro and Ultra, and it needs no card.

The Bottom Line

Gemini 4 Argon is Google's top model and the first of Gemini 4. For most people it costs $19.99 a month through Google AI Pro, which gets it in the Gemini app and Search but not in AI Studio or Antigravity. The free plan and AI Plus don't have it. Developers pay $2 and $10 per million tokens until 30 November and, if Google keeps to its pricing page, twice that afterwards. That would turn Argon's best number on Artificial Analysis, matching GPT-6 Astra for much less per task, into a closer call. Before you buy a plan for it, try it on a real job next to the model you already use.

---

Want Gemini 4 Argon in the same thread as ChatGPT and Claude? It's on every paid izzedo chat plan, from $6 a month, and the free plan needs no card.