What Are Tokens? Why LLM Providers Charge Per Token
A token is the unit a language model reads and writes. It is a chunk of text: a short word, part of a longer word, a number or a punctuation mark.
Tokens vs words
In English, one token is roughly three quarters of a word. One million tokens is about 750,000 words, which is around a 1,500-page book. Code, numbers and unusual words use more tokens per character.
Languages matter
Tokenizers are tuned mostly for English. Text in other languages often needs noticeably more tokens for the same meaning, so the same message can cost more to send. If you serve non-English users, measure real token counts instead of guessing.
Why input and output cost different amounts
Input tokens are processed in parallel. Output tokens are generated one at a time, which uses more compute. That is why output usually costs three to six times more than input.
Other things that count as tokens
- System prompts and instructions you send on every request
- Conversation history in a chat
- Documents and tool results you attach
- Hidden reasoning tokens, billed as output on many models (see hidden LLM costs)
A concrete feel for token sizes
As a rule of thumb, one token is about four characters of English text, or three quarters of a word. A short sentence is around 15 to 20 tokens. A page of text is roughly 500 to 700 tokens. A 10,000-word report is about 13,000 tokens. Numbers, URLs and code break into many small pieces, so a block of JSON often has more tokens than its length suggests.
Context window: the token budget per request
Every model has a context window, the maximum number of tokens it can handle in one request, counting both your input and its output. Current flagship models offer roughly one million tokens, enough for several books. A bigger window does not mean you should fill it: you pay for every token you send, and some models charge more for very long prompts. See long-context price cliffs.
Max output is a separate limit
Models also cap how many tokens they can generate in one reply. Smaller models often have lower caps. If you need long outputs, such as a full report, check this limit before you pick a model.
How tokens turn into your bill
Your cost per request is input tokens times the input price plus output tokens times the output price. Because the unit price is per million tokens, the cost of one short message is tiny. The bill grows from volume: thousands of requests a day, each carrying a long system prompt and history. Reducing tokens that you send on every call is the most reliable saving.
Why the same text costs different amounts
Each provider uses its own tokenizer. The same paragraph might be 1,000 tokens on one model and 1,150 on another. When you compare prices, a lower price per token on a model that uses more tokens can cancel out. For a fair comparison, count tokens for each model on a sample of your real prompts.
How to count tokens
- Read the usage field in the API response
- Use your provider's token-counting tool or library
- Check the usage dashboard for daily totals
- For planning only, estimate from word count
Frequently asked questions
How many tokens is 1,000 words?
About 1,300 tokens in English.
Are spaces and punctuation tokens?
They are part of tokens. Punctuation often takes its own token, and spaces are usually attached to the following word.
Do images and files count as tokens?
Yes. Images and documents are converted into tokens and billed as input, and the count depends on size and the provider's method.
What is the cheapest way to reduce tokens?
Shorten the instructions you repeat on every request, and send only the history and documents that matter. Then calculate the new cost.
Putting it to use
Once you know your average tokens per request, calculate your cost or plug the numbers into the calculator.
Track spend by team and project: the AI Spend Tracker spreadsheet is launching soon.