Google Gemini: I've been hit with a huge bill, what should I do?

First, work out which Gemini you're paying for

Google sells Gemini in two very different ways. Gemini in Google Workspace is a per-person subscription, so a bigger bill usually means more licences were added. The Gemini API, used through Google AI Studio or Vertex AI on Google Cloud, is paid by usage, so a big bill means a lot of tokens were used.

Check the invoice. A Workspace bill lists users and licences. A Google Cloud bill lists projects and services, such as the Generative Language API or Vertex AI.

Stop the spend straight away

If it's usage-based, open the Google Cloud console and find the project with the cost. If you don't recognise the activity, disable or restrict the API key straight away and create a new one for the tools that genuinely need it. Leaked keys, for example ones committed to a public code repository, are a common cause of runaway bills.

If the spend comes from your own tool or automation, pause it. A loop that retries on errors, or a script sending a whole document with every request, can use millions of tokens overnight.

Find out exactly what happened

Use the billing reports in Google Cloud to break the cost down by project, service and day. Look for the day it started and match it to anything that changed: a new automation, a new team member's script, or a key shared outside the business.

Write down the timeline and what you've already done to stop it. You'll need it for the next step.

Ask Google for a billing review

Contact Google Cloud billing support through the console and explain what happened, especially if the usage came from a leaked key or a clear mistake you've since fixed. Google reviews these case by case. There's no guarantee of a refund or credit, but a clear timeline and evidence of the fix give you the best chance.

For Workspace, check with your admin which licences were added and whether they can be removed at the end of the current commitment. [Confirm the current Workspace licence terms before publishing.]

Make sure it can't happen again

Set a budget with alerts in Google Cloud Billing so the right people get an email well before costs climb. Budgets warn you, but on their own they don't stop usage, so also set quotas on the Gemini API to cap requests per minute or per day.

Restrict every API key to the services and apps that need it, keep keys out of code and shared documents, and review usage monthly. Then look at token optimisation: shorter prompts, caching and the right model for each job will lower the normal bill too.

How Gemini API cost is worked out

The Gemini API uses token based pricing. You pay for every input token you send, such as instructions, documents, images and chat history, and every output token the model writes back. Google quotes Gemini API pricing per million tokens, with different rates for each model. Pro models cost the most, Flash models sit in the middle and Flash-Lite models are the cheapest. Rates change as newer models arrive, so check [current Gemini API pricing page] rather than relying on old figures.

Several things can make the bill higher than the headline token price suggests. Thinking tokens are the model's reasoning before it answers. You may not see them, but they're billed as output, so a thinking model on a hard task can use far more tokens than the reply you read. Text input, input image, video and audio input are each priced differently. Very long prompts can move into a higher price band on some models, and features such as grounding with Google Search can carry their own charge. [Confirm current long context and grounding charges before publishing.]

Free tier vs paid tier: what changes when billing is on

Google AI Studio lets you try the Gemini API on a free tier with lower rate limits. Once you link a Cloud Billing account, your project moves to a paid tier and usage costs start to apply. Many surprise bills start here: a project that was tested for free gets billing enabled, and everyone assumes it's still free.

Paid projects are placed in a usage tier, and higher tiers allow more requests. Your tier depends on your billing history and total spend. Rate limits are measured in requests and tokens per minute and per day, and they're set per project, not per API key. [Confirm current tier rules and whether the free tier is available to UK users before publishing.]

Free tier data may also be handled differently from paid tier data, including how it's used to improve Google products. If you send customer or staff information, use a paid project and check Google's current terms for the Gemini developer API.

Why Gemini API cost climbs faster than you expect

Gemini API cost is usually driven by a few things you can see in your usage report. The first is volume of input tokens. Each Gemini model has a large context window, which is useful, but it also means a tool can send a very long prompt without any error to warn you. You pay the base input price for every token in it, every time.

The second is thinking tokens. Gemini models that reason before they answer can produce many more output tokens than the visible reply. On simple jobs that's money spent for no gain. Setting a lower thinking budget, or using a model without thinking for routine work, keeps token costs in check.

The third is how often the same model is called with the same context. If a support assistant sends the same instructions with every question, context caching lets later requests read that part at a lower rate. Batch processing does the same for work that can wait. Both sit alongside standard pricing rather than replacing it, so check which jobs qualify.

Larger organisations on Google Cloud can also negotiate committed use discounts. For most small businesses the bigger savings come from sending fewer tokens and choosing the right Gemini models, priced per million input tokens and per million output tokens.

What to check first in the console

Open the Google Cloud project that holds your API key and check its billing status, so you know it's linked to the account you expect. Then look at cumulative spend for the month against your budget, and at Gemini API usage by model and day. If your account uses prepay billing, check whether you have a negative balance. If the project was meant to be on free access, confirm whether free tier usage limits still apply or billing was switched on.

The usual causes of an unexpected Google Cloud bill

When we look at a big Gemini bill, the cause is usually one of a handful of things.

• A leaked API key, often committed to a public code repository or pasted into a shared document
• An automation stuck in a retry loop, sending the same request again every time it fails
• A script sending a whole document or long chat history with every request
• A thinking model used for simple jobs that a Flash-Lite model could handle
• Image, video or audio input processed in bulk without anyone checking the cost first
• Several projects linked to one billing account, so a test project's spend lands on the main invoice
• A tool that was on the free tier until billing was enabled, then kept running at the same volume

Understanding your Cloud Billing account

Every Gemini API project on the paid tier is linked to a Cloud Billing account. One billing account can pay for many projects, and each project can use many Google Cloud services. The invoice is for the billing account as a whole, so the first job is to see which project and which service the cost came from.

In the Google Cloud console, open Billing, choose the selected Cloud Billing account and go to Reports. Group the cost by project, then by service, then by SKU, and set the date range to cover the spike. This billing data shows you the day the spend started and which Gemini models it came from.

Google AI Studio usage also has its own billing view. If you can see costs in AI Studio that aren't yet in the Cloud console, it may be a timing difference. Billing information can take a while to appear, so don't assume the latest figure is final.

Postpay and prepay billing

Some accounts are billed after use at the end of each billing cycle. Others use prepay billing, where you buy credit first and usage draws it down. With prepay, usage can stop when credit runs out, and in some cases the balance can briefly go negative. Check which you have, because it changes how fast a runaway job can spend. [Confirm current prepay rules and minimum amount before publishing.]

Credits and trials

Promotional credits or a Google Cloud free trial may not cover Gemini API usage in the way you expect. Read the terms of any credit before relying on it to absorb a spike.

How to set a Google Cloud billing budget

A Google Cloud billing budget sends you alerts when actual or forecast spend reaches levels you choose. It is quick to set up and is the first control every business should have.

Create and name the budget

In the console, go to Billing, then Budgets and alerts, and create a budget. Give it a clear name, such as Gemini API monthly.

Set the scope

Choose which projects and services the budget covers. You can scope it to the whole billing account or to a single Gemini project, which makes alerts more useful.

Set the amount and thresholds

Set a monthly amount based on what you expect to spend, then add alert thresholds at several points below and above it. Include a forecast threshold so you're warned when Google predicts you'll go over, not just after you have.

Choose who gets the alerts

Send alerts to billing admins and to the person who owns the tool. An alert nobody reads is no protection.

Spend caps, Gemini API quota and rate limits

Budgets warn you. They don't stop usage on their own. To actually cap spend you need limits that block requests.

Google offers spend caps for the Gemini API at project level, and limits that apply across a billing account depending on its tier. A project cap stops further paid usage on that project once costs reach the amount you set, and each request fails with a tier limit or cap error until the cap is raised or the period resets. [Confirm current spend cap options and how quickly they take effect before publishing.]

You can also lower the Gemini API quota for a project in the console, for example capping requests per minute or per day well below the default. That turns a runaway loop into a short blip. Restrict every API key to the Gemini API only, and to the apps or IP addresses that use it.

For most small businesses, the right setup is a budget with alerts on the billing account, a spend cap on each Gemini project, a sensible quota and restricted keys. Together they cover warning, limiting and preventing.

How to lower your normal Gemini API cost

Once the emergency is over, it's worth looking at the everyday bill. The same techniques that cut costs on any AI provider apply here.

• Use the smallest model that does the job: Flash-Lite models for sorting and tagging, Flash models for most work, Pro only where it's needed
• Control thinking: set a lower thinking budget on simple tasks so the model doesn't over-reason
• Use context caching for long documents or instructions you send many times
• Use batch processing for jobs that can wait, which Google prices lower than standard requests
• Ask for shorter, structured answers to cut output tokens
• Trim chat history and send only the relevant section of a document

Get help with a Gemini bill

If you're not sure where the spend came from, or you want controls set up properly, we can help. We'll read your billing data, find the cause, set budgets, caps, quotas and key restrictions, and then look at the token optimisation that brings the normal bill down. AI finds the patterns in your usage, a person who knows Google Cloud checks every change, and your own data shows what's really happening.