What is an AI token?
A token is one of the small text units an AI model reads and writes. Providers bill by token count because models do not process language in neat human word blocks.
That is why a prompt that looks short to a person can still carry a meaningful token bill, especially once you include hidden system instructions, retrieval context, and tool output.
Token is not word
A token can be a whole short word, part of a longer word, punctuation, or a code-like fragment. That is why word count and token count are related but not identical.
Language matters
Dense technical text, code, symbols, and non-English content can tokenize differently from plain conversational English.
Output counts too
You pay for what the model returns as well as what you send. Long answers are often a larger cost driver than the prompt itself.
A rough mental shortcut
For plain English, a common shortcut is roughly four characters per token. It is only a shortcut, but it is good enough to understand why a “small” message can still become a real line item at scale.
If you paste a few hundred words into a model, you are usually looking at hundreds of tokens, not a dozen. That is why token-aware budgeting matters once usage climbs.
Simple examples
“Summarize this email” is short in words, but the full email body, the system instruction, and the generated summary all count toward tokens.
A chatbot can also look cheap in single-request terms while becoming expensive in monthly terms if every conversation carries a long running history.
When estimates drift
- Large code blocks usually tokenize less intuitively than normal prose.
- Tables, symbols, and JSON often cost more tokens than they look like.
- Very short or very repetitive text can break simple rules of thumb.
Why tokens matter for cost
Providers typically publish prices per million input tokens and per million output tokens. That means even a low rate can turn into a meaningful bill if your prompt is bloated or your answers are consistently verbose.
The practical move is to estimate token size, multiply by request volume, and compare different models or prompt shapes before traffic makes the decision for you.
Best next step
If you want an approximate count from real text, open the AI cost calculator and use the built-in text estimator.
If you are still working backward from pricing logic rather than text size, pair this page with How AI pricing works.
Maintained by David K. · Last reviewed: July 23, 2026
Model token costs, request volume, and scenario-based monthly spend without guesswork.
Open calculatorCheck monthly payment with taxes, insurance, HOA, and a more realistic ownership picture.
Open calculatorWork backward from income, debts, and down payment into a realistic house budget band.
Open calculatorEstimate panel count, annual production, incentives, and a simple payback range.
Open calculator