How Much Does It Cost to Use Gemini 4 Argon?

2026-10-27

Key Takeaways

Gemini 4 Argon costs $2 per million input tokens and $10 per million output tokens on the Gemini API through November 30, 2026.
From December 1, 2026, the standard price doubles to $4 per million input tokens and $20 per million output tokens.
Cached input is billed at a 95% discount: $0.10 per million tokens now, $0.20 after the promotion ends.
Batch and Flex processing cost half the standard rate, while Priority processing costs 1.8 times as much.
One price applies across the full 1,048,576-token context window, with no surcharge for long prompts.
Argon writes many thinking tokens, which are billed as output, so its real cost per task is higher than its sticker price suggests.
There is no free API tier. Consumers can use Argon through Google AI Pro and Ultra subscriptions instead of paying per token.

Gemini 4 Argon. Image source: Google

Using Gemini 4 Argon through the Gemini API costs $2 per million input tokens and $10 per million output tokens until November 30, 2026. Starting December 1, Google’s list price rises to $4 for input and $20 for output. Cached input, batch jobs and priority traffic each have their own rates, and reasoning tokens count as output, so the final bill depends heavily on how much the model thinks before answering.

Most people outside development teams will never see a token bill. Google includes Argon in its Google AI Pro and Ultra subscriptions, where it powers the Gemini app and AI Mode in Search. Developers and businesses pay per token through the Gemini API or Vertex AI, and that is where the December price change matters most.

Gemini 4 Argon API Pricing at a Glance

Google launched Argon on September 30 for a group of trusted cyber defenders and made it generally available on October 13. The launch pricing works as an introductory offer that ends after November. According to Google, the offer covers only the Argon model meant for public access, so specialized versions sit outside it.

Rate per 1 million tokens	Until November 30, 2026	From December 1, 2026
Standard input	$2.00	$4.00
Standard output (includes thinking)	$10.00	$20.00
Cached input	$0.10	$0.20
Batch and Flex input	$1.00	$2.00
Batch and Flex output	$5.00	$10.00
Priority input	$3.60	$7.20
Priority output	$18.00	$36.00

One detail separates Argon from its predecessor. Gemini 3.1 Pro charged more once a prompt passed 200,000 tokens. Argon keeps a single rate across its entire context window of 1,048,576 tokens, so a 900,000-token prompt costs the same per token as a short one.

Why Argon Can Cost More Than Its Price per Token

Sticker prices only tell half the story. What matters is how many tokens a model spends to finish a job, and Argon spends a lot. Thinking tokens are billed at the output rate, and independent testers at Computing for Geeks found that Argon at its high thinking level produced about three thinking tokens for every visible token in its answers.

Artificial Analysis measured the same pattern across its full test suite. Argon used roughly 62,000 output tokens per task in its Intelligence Index, against about 27,000 for GPT-6 Astra. That pushes Argon’s average cost per task to $1.99 at the introductory price and $3.98 at the standard price.

Model	List price (input / output per 1M tokens)	Approximate cost per index task
Gemini 4 Argon (intro)	$2 / $10	$1.99
Gemini 4 Argon (from Dec 1)	$4 / $20	$3.98
GPT-6.1 Sol	$2 / $10	$0.72
Claude Opus 5.5 (high effort)	$4 / $20	About $1.82
GPT-6 Astra	$10 / $50	$3.26
Gemini 3.1 Pro Preview	Previous generation	$0.67

The cost-per-task figures come from Artificial Analysis. The Claude Opus 5.5 number is based on community readings of its chart rather than a published table, so treat it as approximate. Even so, the lesson is clear: at identical list prices, GPT-6.1 Sol finishes the same benchmark work for less than half of Argon’s introductory cost, because it writes far fewer tokens.

Other Charges to Factor In

Caching works automatically, with limits

Argon supports implicit caching only. When the start of a prompt repeats and is at least 4,096 tokens long, Google bills that repeated part at the cached rate without any setup. Explicit cache objects through the cachedContents interface are not supported on Argon, so developers cannot pin a large document in cache on purpose. Keeping system prompts and reference files at the very beginning of each request is the practical way to benefit.

Search grounding is billed separately

Grounding answers with Google Search comes with 5,000 free prompts per month, shared across all Gemini 3 and newer models on the same account. After that, Google charges $14 per 1,000 grounded prompts.

No free API tier

Unlike older Flash models, Argon has no free API quota. Free-tier keys have a quota of zero for this model and return an HTTP 429 error. Anyone testing Argon through code needs a billing-enabled project. Readers who want to use generative AI without a subscription will need to look at other models.

Vertex AI adds a regional premium

On Vertex AI, Argon carries the same list price on the global endpoint, while regional endpoints cost 10% more. Google applies the introductory discount on Vertex as a 50% credit rather than a lower list price.

What Real Requests Cost: Worked Examples

Token prices become easier to judge with concrete numbers. Take a request with 100,000 input tokens that produces 20,000 output tokens, including thinking. At the introductory price, the input costs $0.20 and the output costs $0.20, for a total of $0.40. From December 1, the same request costs $0.80.

Now assume 80,000 of those input tokens are a repeated prefix that hits the implicit cache. At the introductory rate, the 20,000 fresh input tokens cost $0.04, the 80,000 cached tokens cost less than a cent, and the output still costs $0.20, so the total falls to about $0.25. Output dominates the bill, which is why controlling thinking matters more than trimming prompts.

Scale changes the picture quickly. An app that sends 1,000 such uncached requests a day for 30 days would pay about $12,000 a month at the introductory price and about $24,000 after December 1. Moving the same work to the Batch tier would halve either figure, as long as the results can wait.

How to Use Gemini 4 Argon Through a Subscription

Consumers reach Argon through Google AI Pro and Google AI Ultra plans, which unlock it in the Gemini app and in AI Mode. Ultra subscribers also get it in Antigravity, Google’s agentic coding environment, and in the Google AI Studio web interface. Businesses get access through Gemini for business plans and the Gemini Enterprise platform.

As listed in early October, Google’s consumer tiers were AI Plus at $4.99 a month, AI Pro at $19.99, and AI Ultra at $99.99 or $199.99, depending on the usage allowance. Prices vary by country and can change, so check Google’s plan page before subscribing. For a side-by-side look at AI subscription prices across Gemini, ChatGPT, Claude and Grok, our comparison covers the main plans.

Google is also giving some groups access at no charge. It offered 10,000 scientists the top Antigravity tier free for six months, and partners in its Fairwind program for cyber defense receive a separate, more permissive Argon variant for security research.

How Argon’s Price Compares With Rival Models

Model	Input per 1M tokens	Output per 1M tokens
Gemini 4 Argon (from Dec 1)	$4.00	$20.00
Claude Opus 5.5	$4.00	$20.00
Meta Muse Wonder	$4.00	$20.00
GPT-6.1 Sol	$2.00	$10.00
GPT-6.1 Terra	$1.00	$5.00
GPT-6 Astra	$10.00	$50.00
Claude Fable 5.1	$10.00	$50.00
Gemini 3.8 Flash	$0.75	$3.75

After December 1, Argon lands at the same list price as Claude Opus 5.5 and Muse Wonder. Gemini 3.8 Flash remains Google’s budget option, though its price is also set to double on January 1, 2027, to $1.50 input and $7.50 output.

Technical Limits That Affect Your Bill

Argon accepts text, images, video, audio and PDF files, and its knowledge runs to March 2026. A single synchronous request can return up to 65,536 output tokens, although the model’s own API entry lists 32,768, so it is worth testing before relying on the higher figure. For very long outputs, Google’s long decode feature continues generation in 64,000-token steps, up to a total of one million output tokens. Every one of those tokens is billed at the output rate, so long decode can get expensive fast.

Thinking runs at three levels: low, medium and high. Medium is the default. Dropping to low on simple tasks is the most direct lever for cutting costs, since tokens are the real currency of every model bill.

How to Keep Gemini 4 Argon Costs Down

Teams planning to adopt Argon have a clear window: anything run before December 1 costs half as much. Beyond timing, the biggest savings come from sending work that can wait to the Batch or Flex tiers, keeping long shared prefixes at the start of prompts so the implicit cache kicks in, using low or medium thinking unless a task truly needs depth, capping output length, and sending routine jobs to a cheaper model such as Gemini 3.8 Flash. Tracking cost per completed task, rather than price per token, gives the most honest view of what Argon is worth for a given workload.

If you are interested in this topic, we suggest you check our articles:

Sources: Google – Introducing Gemini 4 Argon, Google – Gemini 4 Argon, Computing for Geeks, eesel AI

Written by Alius Noreika