The simple definition
A token is a small chunk of text that an AI model reads, processes and writes. When you type a prompt, the model does not see words the way you do. It breaks your text into tokens first. Everything it generates is also produced one token at a time. Tokens are the basic units of AI language.
Think of a token like a syllable in a jigsaw puzzle. Each piece carries a little meaning. Put the pieces in the right order and you get a sentence. AI models like ChatGPT, Claude and Gemini all work with tokens behind the scenes.
A worked example
Let us tokenise a short sentence: “AI learns quickly.” The model might split it like this:
AIlearnsquickly.
Notice that words are not always whole tokens. Common words often survive intact. Rare words get split into parts. Punctuation usually gets its own token. A word like “unbelievable” might become three tokens: un, believ and able.
Different models use different tokenisers, so the same sentence can split in different ways. But the idea is the same everywhere. The model reads your tokens, predicts the next token, then the next, until the answer is complete. This step-by-step prediction is the core of how ChatGPT works, and it explains why answers sometimes trail off or go wrong.
Why tokens matter
You will rarely think about tokens day to day, but they control three things that affect every user: memory, cost and speed.
1. Memory: the context window
Every AI model has a context window. That is the maximum number of tokens it can consider at once. Your prompt plus the whole conversation plus its answer must fit inside it. When a chat gets too long, the model starts to “forget” the earliest messages. That is not a bug. The oldest tokens simply fall out of the window.
Larger context windows mean longer documents and bigger conversations. This is why new models advertise bigger windows: it is a direct measure of how much they can hold in working memory. If you want the full picture of these models, our guide on what a large language model is explains how tokens fit into the bigger system.
2. Cost: tokens are the billing unit
If you pay for AI through an API, you pay per token. Input tokens cost less than output tokens. Pricing is usually quoted per million tokens, and typical rates are a few dollars or less per million. That sounds cheap, and it is, until you process millions of tokens a day in an automated workflow.
For subscription users the cost is hidden but real. Message limits on free plans are really token limits in disguise. Long prompts and long conversations consume your quota faster than short ones.
3. Speed: fewer tokens, faster answers
Because models generate one token at a time, longer answers take longer to appear. A tighter prompt usually gives you a faster answer and a cheaper one. This is a practical reason to write clear, short prompts instead of rambling ones.
Tokens vs words
| Comparison | Words | Tokens |
|---|---|---|
| What it is | What you and I read | What the model reads |
| Size | Whole units of meaning | Often smaller than a word |
| Average English | One word | Roughly three quarters of a word |
| Common example | “quickly” is one word | “quickly” can be two tokens |
The rough rule of thumb: in English, one token is about three quarters of a word. So 100 tokens are about 75 words. Code and non-English text use more tokens per word, because the tokeniser has fewer patterns to match against.
FAQ
How many tokens are in a word?
In English, roughly one and a third tokens per word on average. Short common words are often one token each. Long or unusual words split into several.
Do tokens cost money?
Yes, when you use an AI API. You pay per token processed, with separate rates for input and output. On subscription plans the cost is bundled into the monthly price, but usage limits still depend on tokens.
What is the context window?
The context window is the maximum number of tokens a model can process in one go. It includes your prompt, the conversation history and the model’s answer. Anything older than the window is effectively forgotten.
Why do AI models use tokens instead of words?
Words would be a much larger and messier vocabulary. A token vocabulary of tens of thousands of pieces lets the model handle new words, typos, multiple languages and code with a compact system it can train efficiently.
Can I reduce the number of tokens I use?
Yes. Write shorter prompts, remove unnecessary words, and start a fresh chat instead of continuing a very long one. For developers, stripping whitespace and trimming context also cuts token counts.
Conclusion
A token is simply the unit an AI model thinks in. Your text goes in as tokens and its answers come out as tokens. Tokens decide how much a model remembers, how much it costs and how fast it responds. Once you understand tokens, message limits and context windows stop feeling like mysteries and start looking like what they are: budgets you can manage.
