AI Token Counter
Paste any prompt or document to estimate its token count and see how much of each major model's context window it would occupy.
Everything runs in your browser. Your text is never uploaded or stored.
0
tokens, estimated
0
words
0
characters
| Model | Context window | Used |
|---|---|---|
| GPT-5 | 400,000 | 0% |
| GPT-4o / GPT-4.1 | 128,000 | 0% |
| Claude (Sonnet / Opus) | 200,000 | 0% |
| Gemini 2.5 Pro | 1,000,000 | 0% |
| Llama 4 | 128,000 | 0% |
| DeepSeek | 128,000 | 0% |
Counts are estimates based on average characters per token, typically within about 15% of the exact tokenizer. Each model tokenizes slightly differently.
What tokens are
Language models read text as tokens, not words. A token is a frequent chunk of characters: common English words are often one token, longer words split into several, and languages like Chinese or Japanese typically use about one token per character. Pricing, rate limits, and context windows are all measured in tokens, which makes the count worth knowing before you send a long prompt.
How the estimate works
The counter uses character-class heuristics: roughly four Latin characters per token, about one token per CJK character, cross-checked against the word count. For typical English prose the estimate lands within about 15% of the exact tokenizer figure. Each model family tokenizes a little differently, so treat the number as a planning estimate rather than a billing-exact figure.
Context windows compared
The table shows how much of each model's context window your text would use. A long report that fits comfortably in a large window can overflow a smaller one, and the answer the model writes needs room in the same window too. Keeping prompts under about half the window leaves space for a full response.
Frequently asked questions
How accurate is the estimate?
For English prose it is usually within about 15% of the exact tokenizer count. Code, dense punctuation, and rare words tokenize less predictably, so give yourself margin on hard limits.
Why do different models count different totals for the same text?
Each model family uses its own tokenizer vocabulary. The same sentence can be 100 tokens in one model and 110 in another. This tool shows one estimate against each model's context size.
How many words is 1,000 tokens?
For ordinary English, roughly 750 words. Text with lots of code, numbers, or non-Latin script shifts the ratio.
Does the counter work for Chinese, Japanese, and Korean text?
Yes. CJK characters are counted at roughly one token each, which reflects how most current tokenizers treat them.
Is my prompt sent anywhere?
No. Counting happens in your browser and the text never leaves your device.
More tools
AI Text Cleaner
Remove em dashes, curly quotes, Markdown symbols, and hidden characters from ChatGPT and other AI text.
Invisible Character Remover
Detect and delete zero-width spaces, non-breaking spaces, and other hidden Unicode characters.
Markdown to Word Converter
Turn ChatGPT Markdown into a formatted Word document, or copy it straight into Word and Google Docs.
Markdown Table to Excel
Extract Markdown tables and export them as CSV, or copy them straight into Excel and Google Sheets.