Reducing token usage: keeping cloud AI costs near zero, on purpose

When a cloud AI service does charge money, it almost always charges by the "token." Understanding what that word means is the difference between a surprise bill and a service you use deliberately, only when it's worth it.

What a token actually is

A token is roughly a small piece of a word — sometimes a whole word, sometimes just part of one. As a rough rule of thumb, 100 words of ordinary English comes to about 130 tokens. Cloud AI services often charge based on how many tokens you send in your question and how many they send back in the answer.

This is why a short, focused question costs a fraction of a cent, while pasting in a fifty-page document and asking for detailed analysis can add up — especially if you do that often.

The good news

Because a local model (see Chapter 02) runs on your own computer, it has no token cost at all — you already paid for the computer. Token costs only apply when you choose to use a cloud-based frontier model. The goal here isn't to avoid those models, just to use them on purpose.

Five habits that keep costs low

Do the first draft locally. Write, organize, and rough out an idea with your free local model. Bring only the polished, specific question to a paid cloud model if you still need one.
Ask once, ask well. A single clear, well-thought-out question usually costs less — and works better — than five vague, back-and-forth attempts.
Don't paste more than you need. If you're asking about one paragraph of a long document, share that paragraph, not the whole document.
Save good answers. If a cloud model gives you a great response, keep it in a personal notes file. There's no need to ask the same question again later.
Reserve cloud models for what they're actually better at — very current information, unusually complex reasoning, or specialized knowledge your local model doesn't have. Save routine work for your free local setup.

A simple way to decide which to use

Use your local model when...

The task is routine writing, brainstorming, or organizing; the material is private; or you're just thinking something through and don't need the absolute best possible answer.

Consider a cloud model when...

You need very recent information, a genuinely hard problem solved, or a capability your local model doesn't have — and you're comfortable with that specific information leaving your computer.

A note on subscriptions

Many cloud AI services offer a flat monthly fee instead of per-token pricing. That can be worth it if you use one regularly — but it's still worth asking, each time, "could my local model have handled this?" Habits, not willpower, are what actually keep a bill small.

Used this way, cloud AI stops being a monthly worry and becomes an occasional, deliberate tool — exactly the role it should play in a private, local-first way of working.