What does AI cost when you bring your own key?

With your own key, you pay the AI provider directly for what your team uses, usually per token, for both what you send and what the model writes back. Your bill depends on the model you choose, how much text each request carries, and how often people use it.

A token is a small unit of text, usually a piece of a word. Bring your own key, or BYOK, means a business tool runs its cloud AI on your company's account with a provider, so usage lands on your provider bill instead of inside a software plan. Our guide to what bring your own key means explains the setup. This guide is about the bill.

How AI providers bill API usage

• Tokens. Models read and write text in tokens, pieces of words a few characters long on average in English. A short email is a few hundred tokens; a long contract can run to many thousands.
• Input and output. The text you send (your question plus any context) is input. The answer is output. They are priced separately, and output usually costs more per token.
• Per model. Each model has its own rates. Larger, more capable models cost more per token than smaller, faster ones from the same provider.
• Per million tokens. Providers usually quote their rates per million tokens, which makes small requests look cheap and heavy use add up quietly.
• Extras. Some providers discount repeated input (often called caching) or requests sent in a batch, and bill images, audio or web search separately.

Rates change often, so read the provider's own pricing page on the day you decide, never a figure from an old article.

What drives the bill, and how to keep it down

Driver	Why it costs	How to keep it down
Model choice	Larger models charge more per token	Use a smaller model for routine work; save the large one for hard questions
Context sent	Every request includes the records, documents and history you attach	Send what the task needs, not everything you have
Conversation length	In a chat, each new message usually resends the conversation so far	Start a fresh conversation for each new topic
Answer length	Output is the more expensive side of the bill	Ask for short answers and bullet points
Volume	More people asking more often	Set a spending limit and look at usage weekly

The third row surprises many teams. A chat that has run for an hour sends that whole hour as input with every new question, so the tenth message can cost many times what the opening message did.

Estimating a month of usage

You can size a budget before you connect anything, with four numbers:

1. How many people will use AI.
2. How many requests each person makes on a working day.
3. How many working days in the month.
4. How many tokens a typical request uses, input and output together.

Say 5 people make 20 requests a day for 21 working days: that is 2,100 requests. At around 3,000 tokens per request, the team uses about 6,300,000 tokens in the month (sample data). Split that into input and output, since they are priced differently, and multiply each by the current rate for the model you plan to use.

Then check the estimate against reality. After the opening week, the provider's usage page shows what you actually used, and you can adjust the model or the habits before the month closes.

Keeping spending under control

• Set a limit. Many provider consoles offer usage dashboards, and some offer spending limits or alerts. Set up what yours offers on the day you create the key.
• One key per purpose. A separate key for each tool or team shows where usage comes from, and lets you revoke one without breaking the rest.
• Pick the model per task. Drafting a polite reminder does not need the largest model you can buy.
• Watch the context. Tools that attach whole records to every request are convenient; check how much they send.
• Review monthly. Compare usage with the value you got. If a use case costs more than the time it saves, drop it.

Your own key or AI bundled into a plan?

Some software includes AI in its subscription price. That makes the bill predictable, but you don't choose the provider and you usually pay for AI whether or not you use it. With your own key, you pay for exactly what you use, at the provider's rates, and you choose who processes your data. Occasional users often pay less with their own key; heavy users should compare their usage estimate with what a bundle includes. How software prices seats and plans matters here too, and our comparison of per-user, per-company and flat-fee pricing covers it.

There is also a third option for some tasks: AI that runs on the device, such as Apple Intelligence, has no per-token bill at all. Our guide to Apple Intelligence for business explains where it fits. And if you are still weighing providers, start with choosing an AI provider for business data.

Frequently asked questions

How much does the OpenAI API cost for a small business?

It depends on the model and on how much your team uses it, and the rates change. Check the provider's own pricing page, then use the four-number estimate above with the current rate for the model you pick.

Does a ChatGPT subscription include API usage?

No. A ChatGPT plan is billed separately from the OpenAI API. Using your own key in a business tool runs on API usage, billed and invoiced separately.

What is a token in AI pricing?

A token is the unit models use to read and write text, usually a piece of a word. Providers count the tokens in each request and each answer and bill per token, often quoted per million.

Why do output tokens cost more than input tokens?

Providers usually set a higher rate for output largely because generating text takes more computing work per token. That is why asking for shorter answers is one of the easiest savings.

How can I cap what AI costs my team?

Set a spending limit or alert in the provider's console, give each tool its own key, pick smaller models for routine work, and review usage every month.