How to use the token estimator
- Enter the text whose token count you want to know.
- The estimated token count and range appear immediately.
- Enter a price per 1 million tokens to get a rough cost as well.
How the token estimator works
AI models process text by splitting it into pieces called tokens. In English, about four characters make a token on average, while Korean often splits to around one token per character. Each model has a different tokenizer, so the same text gives different token counts, and some models have no public offline tokenizer. This tool therefore uses an estimate that multiplies the counts of Korean, Latin, digit, whitespace and other characters by per-type coefficients and adds them up, and it shows a minimum and maximum range. It is meant for sensing the rough scale, not for exact values, and your text is never sent to a server.
Tokens and cost
Most AI services charge by the number of input and output tokens, usually quoted as a price per 1 million tokens. For example, at 3 dollars per million tokens, 2,000 input tokens cost about 0.006 dollars. This tool estimates only the tokens in your input text, so the tokens the model generates in its answer must be considered separately.
Tips
- Use it to roughly check whether a prompt fits within a model's input limit.
- Enter a price to compare and estimate API costs in advance.
- The same content in English tends to use fewer tokens, which is useful when you want to shorten length.
- If an exact limit matters, check the official token counting feature the service provides.
Notes & limitations
The values from this tool are estimates and may differ from the real token count of a specific model. When you must verify prices or limits exactly, check the provider's official tool and price list.
FAQ
How far is it from the real token count?
It depends on the model and the text, and Korean text in particular can differ a lot, so a minimum to maximum range is shown.
Where can I get an exact count?
Use the official tokenizer each AI service provides or the token counting feature of its API.
Why do English and Korean give different token counts?
Tokenizers are often optimized for English word pieces, so Korean tends to be split more finely.