Free tool
How many tokens is this text?
Paste a prompt, a file or a document. You get a token range from two common rules of thumb and the cost of that range on each model. It is an estimate, not a tokenizer count.
Runs in your browser. Nothing you enter is sent anywhere.
The text stays in this tab. Counting happens as you type; nothing is uploaded.
Estimate
Estimated tokens
143 – 148
- Characters
- 590
- Words
- 107
- Characters ÷ 4
- 148
- Words × 1.33
- 143
| Model | As input | As output |
|---|---|---|
| DeepSeek V4 Flash | $0.000034 – $0.000036 | $0.000138 – $0.000142 |
| Claude Haiku 4.5 | $0.000158 – $0.000162 | $0.000786 – $0.000814 |
| DeepSeek V4 Pro | $0.000188 – $0.000196 | $0.000566 – $0.000586 |
| Grok 4.6 | $0.000314 – $0.000326 | $0.000944 – $0.000976 |
| Grok 4.7 | $0.000314 – $0.000326 | $0.000944 – $0.000976 |
| Claude Sonnet 5 | $0.000314 – $0.000326 | $0.001574 – $0.001628 |
| GPT-6.1 Sol | $0.000314 – $0.000326 | $0.001574 – $0.001628 |
| Claude Sonnet 5.5 | $0.000314 – $0.000326 | $0.001574 – $0.001628 |
| GPT-5.6 Terra | $0.000314 – $0.000326 | $0.001888 – $0.001954 |
| Claude Opus 5.5 | $0.00063 – $0.000652 | $0.003146 – $0.003256 |
| GPT-5.6 Sol | $0.000786 – $0.000814 | $0.00472 – $0.004884 |
For a whole workload over a month, use the cost calculator.
Method
How the estimate works
Models do not read characters or words. They read tokens: pieces of text from a vocabulary that each model family defines for itself. This page does not run those tokenizers. It applies two rules of thumb that hold up reasonably well for ordinary English prose:
- Characters ÷ 4. One token is about four characters of English text.
- Words × 1.33. One token is about three quarters of a word, so a word is about 1.33 tokens. Words are counted as runs of characters between spaces.
The two rules rarely agree exactly, so the page shows both and uses the lower and higher as the range. Costs are that range multiplied by each model’s rate per 1M tokens, then rounded to the nearest $0.000002 (at least $0.000002 on a paid model), which is how the gateway bills a request.
Where the estimate is off
- Code, JSON and logs have many symbols and short fragments and usually come out at more tokens than the range suggests.
- Languages other than English, especially Chinese, Japanese, Korean and other non-Latin scripts, often use noticeably more tokens per character.
- Different model families tokenize the same text differently, so the true count varies by model.
- A real request carries more than your text: system prompt, chat history, tool definitions and images all count as input.
What you are billed for is the token count the provider reports for each request, not this estimate. You can see those exact counts per request in the usage log of the console.