Token Counter for AI Models

Tokens are the basic units into which AI models break down text. A token can be a word, part of a word, or a punctuation mark. The token count determines the cost of API calls and whether a text fits into a model's context window.

For the listed GPT models, the counter uses OpenAI's matching Tiktoken encoding. Claude and Gemini use different tokenizers, so the tool gives an estimate based on the GPT-4 encoding for them. Billing and model-version changes can differ from that estimate.

The token counter supports all common text files: .txt, .md, .csv, .json, .log as well as programming files like .js, .py, .java, .html, .css and many more. Binary files like images or PDFs are not supported.

The token count varies by language. English text typically requires fewer tokens than German text. Languages with different writing systems (e.g., Chinese, Arabic) may require significantly more tokens.

No, all calculations take place entirely in your browser. Your texts are neither stored nor transmitted to servers. The tool works completely offline after loading the page.