Why is my Gemini API bill higher than I expected?

The usual causes are thinking tokens (billed at the output rate), Google Search grounding past the 5,000 free requests a month, a more expensive model than you think, long chat history resent on every turn, and leaked or unrestricted API keys.

How do I check my Gemini API billing?

In Google AI Studio, the Billing page shows a daily cost breakdown per project with a model filter, and the Usage page shows tokens, errors and tool requests. Cloud Billing reports show which SKU the money went to.

How can I cap my Gemini API costs?

Set a monthly Project Spend Cap in Google AI Studio (Spend tab). It has about a 10 minute delay and you pay overages in that window. On Prepay billing, a $0 credit balance also stops all keys on the billing account with HTTP 402.

Why is my Gemini API bill so high? How to check it and cut it

Q: Are Gemini API prices going up?

For some models, yes. Google's pricing page lists Gemini 3.8 Flash and 3.6 Flash at $0.75 input and $3.75 output per million tokens through December 31, 2026, doubling to $1.50 and $7.50 from January 1, 2027.

Find where the money went, cap it, and plan for the 2027 price change.

Quick answer

Check the AI Studio Billing page by project and model first. The usual causes are thinking tokens billed at the output rate, Google Search grounding past the free 5,000 requests a month, a pricier model than you meant, chat history resent on every turn, and leaked keys. Set a monthly Project Spend Cap in AI Studio, lower the thinking level for easy tasks, and use Batch or context caching for bulk work.

Where to see what you spent

AI Studio, Billing: daily cost per project, filterable by model, from the last 7 days to the whole month.
AI Studio, Usage: requests, tokens, errors, and tool use such as Grounding with Google Search.
Cloud Billing reports: which SKU spiked. Use this when the AI Studio view is not enough.

The usual causes

Thinking tokens. Google's pricing page lists output price "including thinking tokens". A classification call that thinks for 2,000 tokens before answering in 5 is billed for 2,005 output tokens. Check thoughts_token_count on a few real responses.
Search grounding. 5,000 grounded requests a month are free, shared across Gemini 3 and newer models, then $14 per 1,000. One request can run several search queries, and each query is charged.
The wrong model. Gemini 3.1 Pro Preview is $2 input and $12 output per million tokens, several times a Flash model. A default copied from a demo is a common reason a bill jumps.
Chat history. Every turn resends the whole conversation as input. Long chats cost more per message as they go.
Leaked or unrestricted keys. A sudden spike at odd hours, on models you do not use, is a stolen key until proven otherwise. See Gemini API key leaked.

Plan for January 2027

Google lists Gemini 3.8 Flash and 3.6 Flash at $0.75 input and $3.75 output per million tokens through December 31, 2026, and $1.50 and $7.50 from January 1, 2027. If your current bill is fine only because of the 2026 price, do the cost work before the switch, not after the first invoice.

How to cut it

Lower thinking_level for classification, extraction and short rewrites. Gemini 3.8 Flash defaults to medium and goes down to low.
Do not use a tiny max_output_tokens to save money. It counts thinking tokens, so the answer can come back empty and the thinking is still billed.
Use a Flash or Flash Lite model for easy calls and keep Pro for hard ones.
Trim or summarize old chat turns instead of resending all of them.
Use the Batch API (50% off) for anything that does not need an answer now.
Use context caching for a long system prompt or document reused across calls.

cheap_classify.py

from google import genai
from google.genai import types

client = genai.Client()
resp = client.models.generate_content(
    model="gemini-3.8-flash",
    contents="Classify this ticket: refund, bug or question. ...",
    config=types.GenerateContentConfig(
        thinking_config=types.ThinkingConfig(thinking_level="low"),  # default is medium
    ),
)
u = resp.usage_metadata
print(u.prompt_token_count, u.candidates_token_count, u.thoughts_token_count)

Cap it so it cannot surprise you

Since March 2026, AI Studio has a Project Spend Cap: Spend tab, Monthly spend cap, Edit spend cap. It applies with about a 10 minute delay, and batch jobs and agent sessions can run past it. Billing accounts also have a tier cap (Tier 1 is $250 a month, Tier 2 $2,000), and on Prepay billing all keys on the account stop with 402 when credits hit $0.

If you are paying from India

Google's prices are listed in US dollars. Billing accounts in India pay the equivalent in rupees, and tax can be added on the invoice, so your card statement will not match the price page line for line. Two other things catch people out: the $300 Google Cloud free trial credit cannot be used for the Gemini API, and on the free tier your prompts can be used to improve Google's products. Paid tier prompts are not.

Doing this with Prune

Prune is an AI gateway that works with Gemini for standard and streaming calls. Your Gemini key goes into Prune's encrypted vault, and your code only holds a Prune key. Shield gives each Prune key a daily spend cap that counts input and output tokens, including thinking, and blocks before the call reaches Google, so you are not waiting 10 minutes for a monthly cap to notice. Repeated prompts are served from cache, easy calls on Pro can be routed to Flash, and every response shows the model that served it and what it saved.

With a single line of code, Prune cuts your AI bill and protects your team from runaway spend, leaked keys and dangerous calls.

Start free Help center

Have an account? Sign in · PyPI.