What Is an AI Token Counter?
An AI token counter tells you how many tokens a piece of text uses. Tokens are the units that models like ChatGPT, Claude and Gemini read and write. A token is often a short word, part of a longer word, a number group or a punctuation mark.
Token counts decide two things that matter every day: whether your text fits in the model's context window, and how much an API request costs. This tool gives you a fast estimate, the share of a context window you will use, the cost of one request or a whole batch, and a chunk splitter for long documents. Your text stays in your browser.
How to Use the Token Counter
- Paste your text, such as a prompt, a report or a set of product descriptions.
- Pick a context window size and how long you expect the answer to be.
- Enter your model's prices per million input and output tokens, and the number of requests for batch jobs.
- Read the result. The bar turns orange above 80% of the window and red when the text will not fit.
- Split long text by setting a chunk size and overlap, then download the chunks as JSON or plain text.
Tokens, Words and Characters
OpenAI's guide to tokens gives these rules of thumb for English text:
| Text | Roughly |
|---|---|
| 1 token | 4 characters, or three quarters of a word |
| 100 tokens | 75 words |
| 1 or 2 sentences | 30 tokens |
| 1 paragraph | 100 tokens |
| 1,500 words | about 2,000 tokens |
Hindi, Tamil, Chinese, Japanese and many other languages usually need more tokens for the same meaning, and code with lots of symbols does too. Each model family uses its own tokenizer, so the same text can differ by 10% or more between providers. That is why this tool shows a range, not one exact figure.
Tips to Use Fewer Tokens
- Convert web pages to Markdown before pasting them. Raw HTML can use several times more tokens than the same text.
- Cut repeated instructions. Say each rule once, clearly.
- Set an answer length. Output tokens usually cost more than input tokens.
- Chunk long documents into 300 to 800 token pieces with 10 to 15% overlap for search and RAG systems.
- Check the real count in your provider's usage report for billing, since this tool gives an estimate.
Learn More About Context and Tokens
See how models hold information in AI memory vs context window, learn how chunks are used in RAG explained, and get the basics in large language models explained. To shrink a web page before you paste it, use the HTML to Markdown converter.
Frequently Asked Questions
1How accurate is this token count?
It is an estimate based on how modern tokenizers split English text, and it is usually within about 10 to 15% of the real count. Each provider uses its own tokenizer, so for exact billing check the usage numbers in your provider's dashboard.
2Do ChatGPT, Claude and Gemini count tokens the same way?
No. Each model family has its own tokenizer, so the same text gives slightly different counts. The rules of thumb, such as about 4 characters per token in English, hold for all of them.
3Do spaces and punctuation count as tokens?
Yes. Punctuation marks are often their own token, and a space is usually joined to the word after it. Long runs of spaces, line breaks or symbols in code can add up quickly.
4What chunk size should I use for RAG?
Many teams start with 300 to 800 tokens per chunk and an overlap of 10 to 15%. Smaller chunks give more precise matches, larger chunks keep more context. Test a few sizes with real questions.