AI glossary · Beginner · AI platforms & solutions
Token-based pricing
Token-based pricing is a way of charging for AI models used through an API by counting the amount of text sent and received in tokens. Usually, input tokens fed into the model and output tokens the model writes have separate unit prices.
About this term
- Level
- Beginner · Core ideas for first-time readers
- Category
- Infrastructure & using models
- Fields
- AI platforms & solutions
- Added
- 2026-10-12
How it works
Models read and write text by splitting it into pieces called tokens. Price lists from AI companies and clouds usually list separate input and output prices 'per million tokens', and output is often more expensive. Inputs reused through prompt caching and non-urgent bulk (batch) requests may have separate prices. Reasoning models sometimes count the 'thinking' process before the answer as output tokens.
Input tokens and output tokens
Input tokensText fed into the model
- System prompt, question, retrieved documents
- Grows as conversation history builds up
- Can be reduced with caching
Output tokensText the model writes
- Usually a higher unit price
- Grows with longer answers
- For reasoning models, may include the thinking process
공통 Both are based on token counts from the tokenizer
How to get a feel for it
The cost of one call can be put in words as 'input tokens × input price + output tokens × output price'. Multiply this by the number of calls per day and the number of days to get a rough monthly cost. RAG that feeds in long documents every time, or chatbots whose conversation history keeps growing, see input tokens grow quickly. You can check the actual tokens used in the usage field of the API response or the usage dashboard in the console.
Things to know
A low unit price does not mean a low total cost. The same text yields different token counts depending on the tokenizer, and models that write long answers or think a lot produce many output tokens. Korean sometimes takes more tokens than English for the same content, so it is safer to measure with your own real data. Consumer monthly subscription plans charge a fixed fee within usage limits, so they are hard to compare directly with token-based pricing.
Example
Suppose a company builds an internal customer support chatbot using an API. The person in charge checks the input and output prices on the price list, notices that the system prompt and retrieved documents repeat in every request, and turns on prompt caching to cut input costs. This site's news has also covered a reported analysis that the usage value obtained under the same monthly subscription plan varies greatly by model; you need to know what pricing is based on to read such comparisons correctly.