**What Is an AI Token?**
An AI token is a small piece of information that an AI model reads or produces.
A token is not always a complete word.
A short word might be one token. A longer word may be divided into several tokens. Punctuation, numbers and spaces can also contribute to the token count. The exact division depends on the model and language being used.
When you send a question to an AI system, your words are converted into tokens.
The model processes those tokens and generates its response one token at a time.
There are two main types to understand:
* **Input tokens:** The information you send to the AI
* **Output tokens:** The response generated by the AI
Tokens matter for two practical reasons.
First, AI models have a **context limit**. This determines how much information the model can consider at once. A long conversation, document or codebase consumes more tokens and eventually may exceed that limit.
Second, many commercial AI services calculate usage and API costs according to the number of input and output tokens processed.
Images, audio and other media may also be converted into token-like units the model can process.
So when an AI company offers “one million tokens,” it is not offering one million words.
It is offering the capacity to process roughly one million pieces of text or other encoded information.
Humans see sentences.
AI models begin by seeing tokens.