A token is the smallest unit an AI model uses to read, write and bill text. It is not a word or a letter: it is a fragment. In English, a token averages about 4 characters, or roughly 0.75 words. Everything you pay for on an AI API is measured in tokens, input and output.
At a glance
| What it is | The unit the model chops text into |
| Average size | ~4 characters · ~0.75 words |
| Quick reference | 1,000 tokens ≈ 750 words ≈ 1.5 pages |
| How it is billed | By input and output tokens, at different prices |
How text is chopped up
The model does not see words, it sees tokens. Common words are usually a single token; rare ones, proper nouns, long numbers or words in other languages split into several. That is why the same text costs more or less depending on the language: the content does not change, the number of pieces does.
Input and output are billed differently
Every request pays for two things: the tokens you send (input: your prompt, the instructions, the context) and the ones the model returns (output: its answer). Output costs considerably more per token than input, so a long answer weighs more on the bill than it looks.
Why this decides your bill
Your cost does not depend on how many requests you make, but on how many tokens you move. A prompt with a whole document pasted inside can cost more than a hundred short questions. The formula is always the same: tokens × price per token × request volume. That is where the cost per task comes from, the metric that actually matters.
Count your own
Put your input and output tokens and your monthly volume into the token cost calculator and compare across models: you will see in a minute where your spend comes from. If the number surprises you, start with the 7 levers to cut your bill.
Frequently asked questions
How many words are 1,000 tokens? Around 750 words, or about 1.5 pages of running text. It is an average: numbers, proper nouns and less common languages use more tokens per word.
Are input and output billed the same? No. Output costs several times more per token than input, which is why capping answer length is one of the fastest saving levers.
Does the system prompt count? Yes. Everything you send counts as input on every request, including fixed instructions, unless you use prompt caching so you do not pay for them each time.
At e-ficient we measure the cost per task of companies that already run AI in production and tell you how much you would save with each lever at your real volume. The first audit is free if you start a plan with a three-month commitment and returns a diagnosis within 72 hours.