Gemini API pricing: Argon doubles to $4 and $20 in December

How much does the Gemini API cost per model?

The Gemini API costs $0.10 to $4.00 per million input tokens on its text models, and the price you pay depends on which model you call and when you call it. We read every row from Google's Gemini API pricing page on 8 November 2026. All of them are in US dollars per million tokens on the paid Standard tier, for text, image and video input:

Thinking isn't free. Under Argon, Google's page says: "Every token the model processes is counted and billed, including thinking tokens." So a short answer from a model that thinks first can still bill you for thousands of output tokens that you never see. Batch and Flex cost half as much as Standard. Priority costs 1.8 times as much, which puts Argon at $3.60 in and $18 out until December.

Two older models are on their way out. Google marks Gemini 2.5 Pro as deprecated, and on 1 December it will send those calls to 3.8 Flash. Calls to 2.5 Flash go to 3.5 Flash-Lite on the same day. And the newest model isn't on the list at all. Google announced Gemini 4 Carbon on 28 October and said it "will launch at the same introductory price" as Argon, then "$4 per 1M input tokens and $20 per 1M output tokens," per its Carbon post. For now it goes only to a group of cyber defenders.

What do 1,000 Gemini API requests cost?

1,000 requests of 2,000 tokens in and 500 out cost $9 on Gemini 4 Argon today and $18 from 1 December. That's 80% more than Gemini 3.1 Pro Preview, which stays at $10. We worked out each of these from Google's own rates as we read them on 8 November. The 500 output tokens include any thinking, so if a model thinks at length, it will cost you more than this:

Here's the math for Argon. The input side of each request is 2,000 tokens at $2 per million, which comes to $0.004. The output side is 500 tokens at $10 per million, or $0.005. Add the two and you get $0.009 a request. On 1 December both of those rates double, and so does the bill.

Gemini 3.6 Flash costs the same as 3.8 Flash, on both dates. From December, Argon will even cost more than 3.1 Pro Preview does on its long prompts. Newer won't mean cheaper. If you picked Argon for its launch price, it's time to plan the switch.

Is the Gemini API free?

Yes, on most of the Flash and Flash-Lite models, up to their free rate limits. Google lists a free tier for all of its current Flash and Flash-Lite text models, for the three 2.5 models and for the embedding models. Gemini 4 Argon, 3.1 Pro Preview and all of the image models have none. Google's billing docs say "New accounts begin on the Free Tier," and AI Studio "usage remains free of charge unless users link a paid API key."

The catch is your data. Google's pricing page says the free tier's content is "used to improve our products." The paid tier says "not used." That's the trade. If you send customer data, that line alone is a reason to pay.

Google doesn't print the free limits in its docs. Its rate limits page shows the rules for each tier and sends you to AI Studio for the live numbers on each model. The limits count requests per minute, tokens per minute and requests per day, and they apply to a project, not to a key. The daily ones reset at midnight Pacific time. So if a blog quotes you a fixed free number, treat it as a snapshot of what it saw that day.

What does a month of Gemini API usage cost?

A month of 100,000 requests costs $125 on Gemini 3.1 Flash-Lite, $337.50 on Gemini 3.8 Flash and $900 on Gemini 4 Argon, before tax. That's about 3,300 requests a day, each of them with 2,000 tokens in and 500 out. Here's how the bill moves when you change a setting, and what it does when each of the price rises arrives:

The cached row assumes that the same 1,500-token system prompt starts every request. Those tokens bill at the cached rate, which is a tenth of the full one, so three quarters of your input costs almost nothing. You also pay to keep the prompt in the cache for a 30-day month, but at this size that's only $0.54. The total comes to $236.79, about 30% off the plain bill.

Batch halves everything, but it isn't for a live chat window. Google runs your batch jobs when it can, so they suit work that can wait for the morning, such as tagging or summaries. The grounding row sits on top of any model: the first 5,000 grounded prompts in a month are free, and you pay $14 for each 1,000 after that.

What do grounding, images and embeddings cost on the Gemini API?

Grounding with Google Search costs $14 per 1,000 prompts after the first 5,000 in a month, on Gemini 3 and newer. All of those models share the one free pool, and Google Maps grounding works on the same terms. The 2.5 models are still on the old deal, which gives you 1,500 free a day and then charges $35 per 1,000. On a paid tier, Google also caps Search grounding at 1,500 requests a day for each project by default.

None of the image models has a free tier. Maps grounding on the 2.5 models costs $25 per 1,000 prompts after its free daily allowance. And on Argon, Google also bills "internal tool system instructions and function declarations when using grounding," so a grounded call carries more input than the prompt you wrote.

How do you pay for the Gemini API?

You prepay at least $5 through a Cloud Billing account linked in AI Studio. Google's billing docs set a $5,000 ceiling on prepaid credit. Unused credits "expire after 12 months and are non-refundable." When the balance hits $0, calls fail with HTTP 402 until you top up. Keep a buffer.

Each tier comes with a spend cap. On Tier 1, a billing account can spend $250 a month. Once you've paid $100 and it's been 3 days since your first payment, you move up to Tier 2 and a cap of $2,000. Tier 3 needs $1,000 and 30 days. If you hit the cap, Google pauses "service for all projects linked to that billing account" until the 1st of the next month.

The free Cloud trial won't help here. New billing accounts may get a $300 welcome credit, but for accounts opened after 2 March 2026, Google says "you can't use these credits to pay for Gemini API and AI Studio usage." Guides that tell you to spend the $300 on Gemini predate that change.

What does the Gemini API cost next to the rest of your AI stack?

The Gemini API is billed by the token, and it sits apart from any of Google's chat plans. The card for Google AI Pro at $19.99, which we read on 8 November, lists no API tokens at all, though it does raise your Antigravity limits. So if you also test your prompts in three chat apps, you pay for each of those on top. Here's a common stack, with prices we read the same day:

Whizi, which publishes this page, can't cut the first line. It has no API and no API keys, so your app still pays Google by the token. What it can replace is the chat side. Whizi Pro costs $29.99 a month and puts Claude, GPT and Gemini in one chat, with credits per message that vary by model. That's one bill in place of three if you only chat. If you use Claude Code or Codex, keep the plan that includes them, since Whizi has no coding agent.

How can you cut your Gemini API bill?

Pick the smallest model that passes your tests, then batch what can wait and cache what repeats. Today a request on 3.1 Flash-Lite costs about a third of what it costs on 3.8 Flash. Batch halves the bill on any model. Caching a shared prompt took about 30% off our worked month. And cap thinking when the task is simple, since Google bills it as output.

Watch the calendar too. If you're on Argon for its launch price, test 3.1 Pro Preview before 1 December. If you're on 3.8 Flash, try 3.5 Flash-Lite before 1 January. Then set a spend cap for the project on the Spend page in AI Studio. Google calls it experimental, but it's there so that a loop in your code can't drain your balance overnight.