A token, in practice
A short word is often one token. A page is often a few hundred.
The exact number depends on which model is reading. OpenAI publishes an encoding we can run in the browser. The other counts on this site are estimates, and they are labeled that way.
One paragraph, four counts
A token is a small piece of text a model reads or writes. It is not always a whole word. Short words are often one token, and longer words are often split. Spaces and punctuation count too. A page of ordinary English is usually a few hundred tokens, but code, tables, and other languages can land far from that. The numbers beside this paragraph were counted with the same methods the rest of the site uses, so you can see the gap between an exact encoding and an estimate.
89 words. 473 characters.
- OpenAI o200k_base: 100 tokens. This is an exact count for GPT-5 and GPT-6 on this site.
- OpenAI cl100k_base: 100 tokens. This is the exact count for GPT-4 Turbo and GPT-3.5.
- Claude, newer tokenizer: 190 tokens. Estimate from about 2.5 million characters per million tokens.
- Claude, earlier tokenizer: 119 tokens. Estimate from about 750,000 English words per million tokens.
- Gemini: 119 tokens. Estimate from Google’s figure of about 4 characters per token.
- xAI, Mistral, DeepSeek, Cohere: 119 tokens. A 4-character rule of thumb, not a figure those companies published.
The words this calculator uses
Input is the text you send. Output is the text the model sends back. Output is often the expensive part.
Cached input is input the provider already stored, such as a standing instruction. It is part of the input. Some companies charge less to read it again. Writing the cache the first time can cost more, and that write price is not in the main total.
Context window is how much text fits in one request. The bar on the counter shows how full that window is. Near the end, it warns you. Past the end, the request does not fit.
A rough page of ordinary English is a few hundred words. Dense pages, code, and tables can be much more. Paste the page if the bill matters.
How many tokens are in a word?
In English, a short common word is often one token, and a longer word is often split into two or more. A rough average people use is about three quarters of a word per token, but the real count depends on the tokenizer. OpenAI’s count on this site is exact for the encoding we load. The other counts are estimates.
How many tokens are in a page?
A typical double-spaced page of English is a few hundred words, which often lands around 300 to 500 tokens. A dense PDF page, or a page with tables and code, can be much higher. Paste the page into the counter if you need a number you can bill from.
What is a token?
A token is a chunk of text the model reads or writes. It might be a whole word, part of a word, a space, or a mark of punctuation. You pay for input tokens you send and output tokens the model sends back.
What is cached input?
Cached input is text the provider has already stored from an earlier request, such as a long standing instruction. When you send it again, some providers charge a lower read price. The cached tokens are part of the input, not an extra pile on top. Some providers also charge a higher price to write the cache the first time. That write price is not in the main total here.
What is a context window?
The context window is how much text the model can consider at once. If your prompt plus the reply would pass it, the request does not fit. The bar on the counter shows how much of each model’s window your text would use.