Gemini API Pricing Overview: Cost per Million Tokens Across 11 Models

How Gemini API Billing Works

Google bills Gemini API calls by token, split into three rate tiers: input, cached input, and output, all measured per million tokens. For every Gemini model, output pricing is several times the input price, while cached input typically costs only a tenth of input or less. The cost of a request equals the number of tokens consumed multiplied by the corresponding model's rate—there are no seat fees or subscription plans.

This pricing structure means that high-frequency scenarios with highly repetitive input (such as chat services with long system prompts) can benefit significantly from caching, while output-intensive tasks (such as long-form generation) require prioritizing the output unit price.

Tokens are the basic unit that large language models use to process text—neither characters nor words. In English, one token corresponds to roughly 4 characters or 0.75 words; in Chinese, one character typically maps to 1–2 tokens. During inference, input text is first split into a token sequence by the tokenizer, and output is generated one token at a time. So "per million tokens" pricing essentially reflects the model's computational load: input tokens go through a single forward pass, whereas output tokens are generated autoregressively one by one, making them more expensive to compute—this is the fundamental reason output pricing is usually several times that of input. For rough cost estimates, you can use "1 token ≈ 0.75 English words" or "1,000 Chinese characters ≈ 1,500 tokens," but exact values require the model's tokenizer or the official token-counting API.

Full Price List for All 11 Gemini Models

There are currently 11 Gemini models on sale. The table below lists both the prices from third-party aggregator UplinkSwitch and Google's official list prices for easy comparison.

Model	Input	Cached Input	Output	Official Input	Official Cached	Official Output
Gemini 4 Argon	$0.60	$0.06	$3.00	$2.00	$0.10	$10.00
Gemini 3.8 Flash	$0.45	$0.045	$2.25	$0.75	$0.075	$3.75
Gemini 3.5 Flash	$0.90	$0.09	$5.40	$1.50	$0.15	$9.00
Gemini 3.1 Pro Preview	$1.20	$0.12	$7.20	$2.00	$0.20	$12.00
Gemini 3.7 Flash	$0.45	$0.045	$2.25	$0.75	$0.075	$3.75
Gemini 3.6 Flash	$0.45	$0.045	$2.25	$0.75	$0.075	$3.75
Gemini 3.5 Flash Lite	$0.18	$0.018	$1.50	$0.30	$0.03	$2.50
Gemini 3.1 Flash Lite Preview	$0.15	$0.015	$0.90	$0.25	$0.025	$1.50
Gemini 3 Flash Preview	$0.30	$0.03	$1.80	$0.50	$0.05	$3.00
Gemini 2.5 Flash	$0.18	$0.018	$1.50	$0.30	$0.03	$2.50
Gemini 2.5 Pro	$0.75	$0.075	$6.00	$1.25	$0.125	$10.00

The lowest price in the table is Gemini 3.1 Flash Lite Preview: $0.15 input, $0.90 output. Interestingly, several Flash-series models (3.6, 3.7, 3.8) are priced identically, indicating that Google has adopted a uniform pricing strategy at the Flash tier—model selection should therefore be driven by capability differences rather than price differences.

Cost Estimates for Two Typical Request Types

To help illustrate actual costs, the original article assumes two request scenarios: a single conversation turn (2,000 input tokens + 500 output tokens), and a long-document processing request (50,000 input tokens + 2,000 output tokens). Both are calculated at non-cached rates.

Model	Conversation Turn (Official)	Conversation Turn (Here)	Long Document (Official)	Long Document (Here)
Gemini 4 Argon	$0.009	$0.0027	$0.12	$0.036
Gemini 3.1 Pro Preview	$0.01	$0.006	$0.124	$0.0744
Gemini 3.8 Flash	$0.0034	$0.002	$0.045	$0.027

The calculation is straightforward: input tokens multiplied by the input rate, plus output tokens multiplied by the output rate, with rates prorated per million tokens.

Looking at the discount margins: Gemini 4 Argon here is $0.60 input / $3.00 output, versus official $2.00 and $10.00—70% lower on both input and output; Gemini 3.1 Pro Preview here is $1.20/$7.20 versus official $2.00/$12.00—40% lower on both; Gemini 3.8 Flash here is $0.45/$2.25 versus official $0.75/$3.75—also 40% lower.

How Cached Input Saves Money

When a request repeats prefix content the API has already seen (such as a long system prompt or the same document), the response marks those tokens as cached and bills them at the cached rates in the first table. If a model has no cached rate, those tokens are still billed at the input rate.

For RAG applications, customer service bots, document Q&A, and similar scenarios, cached input is a key cost-control lever—cache the fixed, unchanging long prompt and pay only for the actually new tokens each time.

Context Caching works much like CPU caching: if the beginning of multiple requests is identical, the server can save and reuse the intermediate computation (KV Cache) for that portion. Google's Gemini API offers two caching modes—implicit caching, where the system automatically detects repeated prefixes and bills hits at a discounted rate; and explicit caching, where developers proactively create a cache object and specify a time-to-live (TTL), suited to very large documents or system prompts reused over the long term. In addition to charging cached tokens at a lower rate, explicit caching also incurs a cache storage fee based on storage duration, so it only pays off when reuse is frequent enough. Note that caching has a minimum token-length threshold—short prompts won't trigger it; and once cached content changes, all subsequent tokens are invalidated and recomputed, which is why fixed content should go at the front of the prompt and variable content at the back.

Integration and Model Selection Tips

UplinkSwitch offers an OpenAI SDK-compatible interface: just point base_url to its endpoint and fill in your key, and all Gemini model IDs in the table can be called directly:

from openai import OpenAI

client = OpenAI(
    base_url="https://api.uplinkswitch.com/v1",
    api_key="sk-uplink-live-...",
)

completion = client.chat.completions.create(
    model="gemini-4-argon",
    messages=[{"role": "user", "content": "Summarize this in one line."}],
)
print(completion.choices[0].message.content)

On payment: there's no subscription—prepaid credits start at $10, and each top-up locks in pricing for 30 days.

As a selection principle, use the cheapest model that gets the job done: Flash Lite for high-concurrency short-text scenarios, Flash for most chat and coding tasks, and Pro or Gemini 4 Argon for hard problems that need stronger reasoning. The most reliable approach is to run the same prompt across models and compare quality and cost before deciding.

The OpenAI SDK-compatible interface used in the example is a de facto industry standard: as long as the server implements the same request and response structure as OpenAI's /v1/chat/completions, you can reuse the official openai Python/Node packages directly, changing only base_url and api_key. This brings two practical benefits—existing OpenAI-based code migrates with almost no changes, and switching between providers requires changing a single string. However, compatibility layers usually cover only core chat completion; model-specific parameters (such as certain multimodal input formats, particular tool-calling extensions, thinking-budget controls, etc.) may not be passed through. Before integrating, confirm whether the aggregator supports the key capabilities of the model you're using, and verify via actual calls that response fields are complete.

FAQ

Is there a monthly subscription? No—only prepaid credits, with each top-up locking pricing for 30 days.

Which Gemini model should I choose? The cheapest one that gets the job done. Flash Lite suits high-throughput short text, Flash covers most chat and coding, and Pro and Gemini 4 Argon are reserved for hard tasks worth paying more for better answers.

Where's the full price list? All models can be viewed on UplinkSwitch's Gemini pricing page and model pages; to estimate your own usage costs, you can re-price exported usage records on its savings calculator page.

Prices come from UplinkSwitch's live catalog (retrieved 2026-10-17); the original page re-reads the latest prices on each load.

One-line Summary

A breakdown of three-tier token pricing, cost estimates, and selection strategy across 11 Gemini models

Paragraph Summary

The article systematically explains Gemini API billing: three token tiers (input, cached input, output), no subscription, output typically several times input, and cached input about a tenth of input. It then lists official prices for all 11 models on sale alongside third-party aggregator UplinkSwitch's prices—the lowest being Gemini 3.1 Flash Lite Preview ($0.15 input, $0.90 output per million tokens)—and notes that Flash 3.6/3.7/3.8 are identically priced, so selection should be based on capability rather than price. The article also estimates real costs for conversation turns and long-document scenarios, showing long-document requests are the main cost driver, and offers tiered selection guidance (Flash Lite, Flash, Pro/Argon) plus caching optimization tips.