What does this token calculator count?
It counts the tokens in the text you enter, using the o200k_base GPT tokenization rules. It also shows words, characters and whitespace counts. Tokens may contain a word, part of a word, punctuation or whitespace.
Is one character always one token?
No. A token may contain several characters or only part of a word, and one character may use several tokens. Emoji and letters with combining marks can also use multiple tokens even when they look like a single character. There is no fixed character-to-token ratio.
How are Chinese, Japanese and Korean text counted?
All three are supported, but grouping them as CJK does not give them a shared token count per character. Common character sequences may form one token, while a single character or syllable may be split into several. The count depends on the actual text and tokenization rules, not just the language.
What about other languages and mixed text?
The counter also supports text in Latin, Cyrillic, Arabic, Indic and other scripts, including mixed-language text. Words may be split into smaller pieces, so English word or character estimates do not apply to every language. Translations of the same sentence can have different token counts.
How are characters and words counted?
In modern browsers, a Han character, Japanese kana, Korean syllable or Latin letter usually counts as one character. A letter with a combining accent or a combined emoji usually also counts as one. Whitespace and punctuation are included. Words follow language-aware boundaries, so a CJK passage is not treated as one word simply because it has no spaces. Both counts are separate from tokens.
Do spaces count as tokens?
Spaces and other whitespace are included when counting tokens. A space may be combined with the following word or appear in a separate token. Removing a space can change nearby token boundaries, so one space does not always mean one token.
Will this count match every AI model?
No. Models can use different tokenization rules. This tool provides a GPT reference count for plain text. Complete chat requests may also contain roles, message formatting and tool definitions, so their total usage can differ.
Is my text uploaded?
No. Text and imported files are counted in your browser. The tool does not send their contents to a counting service.
Is there a text or file size limit?
There is no fixed text length or file size limit. Large inputs take more time and depend on your device’s available memory. The tool counts the full text and previews only the first 500 tokens; the preview does not reduce the total. Use Clear to stop the current count.