DevOps LoungeToken counter
Paste a prompt, a document or code. See roughly how many tokens it uses, whether it fits common context windows, and what it costs to send.
Fits in a context window of
Rules of thumb
| Text | Roughly |
|---|---|
1 token | 4 characters of English, or about 3/4 of a word |
100 tokens | About 75 English words |
1,000 words of English | About 1,300 tokens |
A page of code | About 500 to 800 tokens |
Arabic or Russian text | Often 1.5 to 3 times the tokens of the same length in English |
How it works
- An honest estimateCommon English words are one token, long words several, and Arabic, Russian and Chinese text use more tokens per character. The range covers how real tokenizers differ.
- Context windowsThe bars show how much of each common window size your text fills, before the model writes a word back.
- CostEnter your model's current input price to see what one request costs. Prices change, so take them from your provider.
Questions
Is this exact?
It is an estimate, usually within 10 to 15 percent. Exact counts need the model's own tokenizer, and every model family has a different one.
Why does Arabic use more tokens?
Tokenizers are trained mostly on English, so other scripts get split into smaller pieces.
Is my text sent anywhere?
No. Counting happens in your browser.
More tools
Built something with it?
Show it in the server, get a code review, or ask the people who use these tools every day.
online now · members