Gemini API pricing: Argon, Flash and Pro per 1M tokens

Google prices the Gemini API per million tokens, with separate rates for input, output and cached input, and thinking tokens bill as output. Argon, the flagship, costs about 2.7 times as much as 3.8 Flash on input and output. Both prices are set to rise: Argon's double on December 1, 2026, and 3.8 Flash's on January 1, 2027.

What does the Gemini API cost per million tokens?#

Google lists a price per million tokens for each model, with separate rates for input, output and cached input. These are the Standard paid-tier prices for the Gemini text models we offer on the API, as Google's pricing page showed them on November 1, 2026. Google still marks 3.1 Pro and 3 Flash as previews.

Four things on that page are easy to miss:

• Two prices are already dated. Argon's current prices run through November 30, 2026, and double from December 1. The prices of 3.8 Flash run through December 31, 2026, and double from January 1, 2027. A cost plan made today should use the later prices.
• Long prompts cost more on the Pro models. On 3.1 Pro and 2.5 Pro, a prompt over 200K tokens is billed at twice the input and cache price and one and a half times the output price. Argon, 3.8 Flash and 3 Flash have one price whatever the prompt length.
• Thinking is output. Every output price includes thinking tokens. On Argon, Google adds that every token the model processes is counted and billed, thinking tokens included.
• Audio costs more. On 3 Flash, audio input costs twice the price of text, image or video.

How is a Gemini API bill calculated?#

Each request adds up from a few parts, each with its own rate:

• Input tokens: the text, images, files and conversation history you send, apart from what is read from the cache.
• Cached tokens: implicit caching is on by default for the 2.5 and newer models. When a request repeats a recent prefix that is long enough, that part is billed at the cached rate: a tenth of the input price on most models, and a twentieth on Argon. The minimum is 4,096 tokens on Argon, 3.8 Flash and 3.1 Pro, and 2,048 on 2.5 Pro. When a request hits the cache, Google passes the saving on automatically.
• Cache storage: explicit caching, where you create a cache yourself, adds a storage charge per million tokens per hour. Argon supports only implicit caching.
• Output tokens: the answer plus the model's thinking.
• Grounding with Google Search: on Google's newer models, the first 5,000 search requests a month cost nothing, shared across those models. After that, Google charges per 1,000 requests, with a cap of 1,500 requests a day. One request can trigger several searches, and each one is charged. 2.5 Pro has its own allowance of 1,500 requests a day, then a higher price.

The processing tier then changes the token prices. Batch and Flex cost half the Standard input and output price on every model in the table, and a batch job targets a 24-hour turnaround. Priority costs 1.8 times the Standard price.

With 5abeer there is no cache field to set for Gemini models: keep the stable part of each request at the start, as Prompt caching explains. Our docs do not list Batch, Flex, Priority or Search grounding, so plan your costs with us on regular requests.

A worked long-prompt example#

Say a request sends a 300,000-token prompt to 3.1 Pro and gets a 10,000-token answer. The prompt is over 200K tokens, so both parts are billed at the higher rates, and the request costs about twice as much as the same request on Argon today. From December 1, when Argon's prices double, the two come out almost even.

A short request changes the picture. With 50,000 tokens in and 5,000 out, Argon is slightly cheaper than 3.1 Pro today, and close to twice the price from December 1. So the cheaper model depends on the date and on how long your prompts are. Run the numbers on your own traffic, and check the cached-token count in the usage data of real replies.

Which Gemini model is cheapest for your work?#

Price per token is only half the decision. A cheaper model that needs three attempts can cost more than a stronger one that gets it right the first time. Google describes the five like this:

• Argon is Google's flagship reasoning model, for complex software engineering, deep math and science problems, and autonomous multi-step agents. Until November 30 it costs no more than 3.1 Pro on any rate.
• 3.8 Flash is Google's most intelligent Flash model, for long-horizon software engineering, autonomous agents and complex enterprise workflows, at a little over a third of Argon's input and output price.
• 3.1 Pro is Google's third-generation Pro model, for multimodal understanding, agentic work and coding. Watch the long-prompt rates above 200K tokens.
• 3 Flash is Google's legacy Flash model, with baseline speed and intelligence, and the cheapest of the five: two-thirds of 3.8 Flash's input price and four-fifths of its output price.
• 2.5 Pro is the oldest of the five, a Pro model that Google says excels at coding and complex reasoning. It costs less than 3.1 Pro on every token rate, and its Search grounding is priced on its own.

A common pattern is to split work by task: a Flash model for the bulk of simple requests, and Argon or a Pro model for the steps that need it. Test quality on your own tasks before you move a workload. For a coding agent in the terminal, see Gemini CLI with 5abeer.

Is there a free tier?#

Yes, on some models. Google's free tier covers 3.8 Flash, 3 Flash and 2.5 Pro. Argon and 3.1 Pro are not on it. Content sent on the free tier is used to improve Google's products, and content on the paid tier is not. Batch is not available on the free tier. For production work, or anything private, plan around the paid tier.

What do the same models cost through 5abeer?#

5abeer sells Gemini models on the API below Google's list price, paid from your balance. The list below reads our current prices from the live catalog, so it stays correct when they change.

Google also lists image and other models that this guide does not cover. The Gemini API page shows what we offer, and each model's page shows its prices and the API formats it supports.

How do you buy Gemini API access through 5abeer?#

1. Create a key for Gemini models in your dashboard. A Claude or GPT key does not work under /v1beta. API keys explains the spend limit, expiry date and IP allow-list you can set on each key.
2. Add balance. API usage is deducted from your balance, token by token.
3. Point your client at our Gemini base URL and send the key in the x-goog-api-key header. The generateContent reference has examples, and Base URLs gives the address for each API format.

To compare Gemini with the Claude and GPT models on the same balance, see the API page. Gemini models also run in apps such as Antigravity. For the app plans, see the apps and pricing.

Questions#

Are thinking tokens billed?#

Yes, as output. Every output price in Google's table includes thinking tokens, so a task that makes the model think longer costs more even when the answer is short.

Will Argon and 3.8 Flash get more expensive?#

At Google, yes. Argon's prices double on December 1, 2026, and those of 3.8 Flash on January 1, 2027. We checked these dates on November 1, 2026.

Does a long prompt cost more?#

On 3.1 Pro and 2.5 Pro, yes: past 200K tokens, input and cache cost twice as much and output one and a half times as much. Argon, 3.8 Flash and 3 Flash keep one price.

Is caching automatic?#

Yes, on the 2.5 and newer models, once the repeated prefix reaches the minimum size: 4,096 tokens on Argon, 3.8 Flash and 3.1 Pro, 2,048 on 2.5 Pro. Put the stable part of the prompt first to raise the hit rate.

Is the Batch API worth it?#

If your requests can wait, it halves the input and output prices at Google, with a 24-hour target turnaround. Our docs do not list it.

How is Google Search grounding charged?#

Per search request, after a monthly allowance shared across Google's newer models, with a daily cap. One prompt can trigger several searches. The prices are in the table above.

Is a subscription cheaper than the API?#

It depends on how much you use and where. A subscription works inside the apps, while the API is paid per token from your balance. AI subscription vs API explains the difference.

Sources#

Checked on 2026-11-01:

• Pricing, Gemini API docs, last updated October 30, 2026
• Context caching, Gemini API docs
• Batch API, Gemini API docs