Is Gemini 4 Argon free? What Google AI Pro, Ultra, and the API cost each month

The day after Gemini 4 Argon shipped, the same three questions showed up together: is it free, is Pro enough, and what does the API cost in a month. Those three bills are rarely on one page. This piece separates the Gemini app subscription from the Developer API, then turns four usage shapes into a monthly dollar figure, including the rate that changes after November 30.

Figures were checked on October 14, 2026 against the Google AI subscriptions page, the Gemini Developer API pricing page, and Google’s October 13 model post and app availability note. Region, tax, and promotions follow the checkout page. For how input and output tokens are counted, see the token pricing comparison.

Two bills, not one

“Is Gemini 4 Argon free?” does not have one answer that covers every door. Google put the same model on two paths:

• The Gemini app and AI Mode in Search — a monthly subscription, with a usage cap. The bill is a fixed fee.
• The Gemini Developer API — model ID gemini-4-argon, priced per million tokens. There is no unlimited monthly plan.

The October 13 launch note is specific. Developers and enterprises use the paid API in AI Studio and the Gemini platform. Google AI Ultra subscribers can try it in Google Antigravity and AI Studio. Everyday use in the Gemini app and in AI Mode starts the same day for Ultra and Pro subscribers. A separate scientist program is an application, not a public free tier.

In the app: who gets Argon, and the monthly list price

The table stays with what the US subscriptions page actually lines up. Plus varies by country and its feature list does not include Argon, so it is not an Argon ticket. Pro at $19.99 is the published US price for that tier. Checkout still follows the account’s country.

The cap is not “unlimited tokens.” The footnote says Gemini app usage depends on prompt complexity, the features you use, and chat length. The allowance refreshes every five hours until you hit the weekly limit. AI credits can extend it. Pro at $19.99 buys capped Argon chat, not an API you can wire into a product.

Read Antigravity as its own line. The subscriptions page gives Pro “entry rate limits” for the agent model, and Ultra higher or the highest limits. The model launch post puts the Argon trial in Antigravity and AI Studio on Ultra subscribers. Those two sentences do not add up to “the Pro fee already includes Argon on a coding-agent API.”

There is also a research lane. Google says it will give 10,000 scientists the highest Antigravity usage tier free for six months, onboarded on a rolling basis, subject to eligibility and capacity. That is an application. It is not a personal account that opens free Argon.

The API: introductory rates through November 30, then December 1

The developer pricing page puts gemini-4-argon on the paid tier. The free column is “Not available” throughout. Standard, Batch, Flex, and Priority are all priced per million tokens in USD, and the output price includes thinking tokens. Explicit caching is not supported. Implicit caching is.

“Cached input tokens at 95% off” means you pay 5% of the input price, not a 5% discount. That matches the pricing page: 5% of the $2 introductory input rate is $0.10, and 5% of the later $4 rate is $0.20. The footnote limits the introductory price to the Argon model intended for public access.

Search grounding is separate. Gemini 3 and newer models share 5,000 free Google Search requests a month, then $14 per 1,000. One user request can fan out into several search queries, and you pay per query. Maps grounding is the same shape: 5,000 free requests a month, then $14 per 1,000 queries.

On the paid tier, prompts and outputs are not used to improve Google’s products. Other models on the free tier still can be. Argon itself is not on that free tier.

Four usage shapes: what a month actually costs

The API has no official monthly bundle. These four rows are worked examples, so they can sit next to the subscription fee. Standard lane, before tax, before search grounding. The December 1 column doubles because both the input and the output introductory rates are half of the later price.

Set that table next to the subscription and a month falls into three answers:

• Argon only in the app. Free is $0 and does not include Argon. Pro is $19.99. Ultra is $99.99 or $199.99. You stop at the weekly cap. The bill does not keep growing per token.
• API only. The table is the bill. Daily coding is about $60 during the introductory window and about $120 from December. A small-team agent with a lot of thinking can already be $120, then $240.
• Both. Pro’s $19.99 is not deducted from the API invoice. Daily coding plus Pro is about $80 before December and about $140 after, before tax.

How thinking, cache, and search make the same reply cost more

Thinking tokens are output

The pricing page says “output price, including thinking tokens.” A short answer on screen can still be billed for the tokens the model spent reasoning. The small-team row puts output at 8 million for that reason: 20 million input is $40, and 8 million output is $80. On a long agent loop, the expensive column is usually output, not how long the prompt was.

Implicit cache discounts input only

Take a long-context month: 40 million input tokens, 30 million of them implicit cache hits, 10 million fresh, plus 4 million output. Introductory Standard is 10M × $2 + 30M × $0.10 + 4M × $10 = $20 + $3 + $40 = $63. With no cache hits it is 40M × $2 + $40 = $120. Cache saves repeated preface. It does not save thinking or the answer.

Search grounding is per query, not per chat

The 5,000 free searches a month are shared across Gemini 3 and newer models. Argon does not get another 5,000 of its own. After that, $14 per 1,000 queries. If one answer fires four searches in the background, that is four queries. Eight thousand queries are about 5,000 free + 3,000 × $0.014 = $42, on top of the token bill.

Priority is a speed premium

The daily-coding row (15M input + 3M output) on Priority during the introductory window is 15 × $3.60 + 3 × $18 = $54 + $54 = $108, not the $60 Standard figure. From December 1 the same row is 15 × $7.20 + 3 × $36 = $216.

Chat, code, or production

• You want to try Argon in the browser or the phone app. Free and Plus do not list it. Pro at $19.99 is the lowest published app door. Stay there if the weekly cap is enough.
• You want long tasks in Antigravity, or a direct try in AI Studio. The launch note puts those Argon trials on Ultra. The gap between $99.99 and $199.99 is rate limit, not the model name.
• You want it inside your own product. Turn on paid API billing and set the model to gemini-4-argon. Send evals and overnight jobs to Batch. Leave live chat on Standard. Put the router in before December so the price change does not require a product rewrite.
• The context and the toolchain are long. A 1 million token unified window raises the input on a single request. Repeated system prompts and repo notes are what implicit cache is for. Do not resend them as fresh input every turn.

If you are still comparing token prices across frontier models, the Claude and Gemini comparison is about an earlier capability band. Do not paste a Flash rate onto Argon.

Easy mistakes

• “The headline says available for everyone, so it is free.” The headline is about a global door. The next sentence splits paid API, Ultra on Antigravity, and Pro or Ultra in the app. The free tier is not in that list.
• “I already pay for Pro, so my API key can call Argon.” The pricing page separates the Pro plan from the Argon API. Seeing the model in the app does not mean gemini-4-argon rides on the subscription fee.
• “The answer is short, so it is cheap.” Thinking tokens sit on the output price. A short reply can hide a long trace.
• “Batch and Standard share one monthly plan and the cheaper rate is chosen for me.” They do not. Batch is another lane, with different latency. You have to send the job there.
• “November’s invoice is next year’s price.” On December 1 the Standard unit prices double. Cache, Batch, Flex, and Priority move with them.

Questions

Is Gemini 4 Argon free?

No. The free Gemini app stays on the Flash-Lite tier, and the Developer API free tier does not offer gemini-4-argon. The app path is Google AI Pro or Ultra. Your own program needs the paid API tier.

How much is Google AI Pro each month?

The US list price is $19.99 a month. The subscriptions page puts Argon on Pro for the Gemini app and for Search, at about 4× the free allowance, with 5 TB of storage and 1,000 Flow credits. That fee does not include the API.

What is the dollar gap between Ultra and Pro?

Ultra starts at $99.99 a month, about 5× Pro usage. The $199.99 tier is about 20×. Storage starts at 20 TB, and Antigravity rate limits are higher. The launch note puts the Argon trial in Antigravity and AI Studio on Ultra.

What does the API cost in a month?

There is no flat monthly fee. Standard through November 30 is $2 per million input tokens and $10 per million output tokens. The daily-coding example is about $60, then about $120 from December 1. The more thinking sits in the output, the closer you are to that $10 column, later $20.

Written on October 14, 2026. Plan benefits were checked on the Google AI subscriptions page. API unit prices were checked on the Gemini Developer API pricing page and the October 13 launch note. The token volumes in the examples are there so they can sit next to a monthly fee. They are not an official bundle.