Count tokens and price a prompt before you send it.

1. 1

   Paste your prompt

   Paste or open the text you will send. The estimate updates as you type.

2. 2

   Set the reply and the volume

   Enter how long the reply will be and how many requests you send each month.

3. 3

   Compare the models

   Read the cost per request and per month for each model. Change any price to match the provider’s page, or add a model of your own.

Language models read and bill text in tokens: pieces of words, numbers and symbols. A short English word is usually one token, a long word several, and code, numbers and non-English text use more per word. Knowing the token count tells you whether a prompt fits in a model’s context window and what it will cost you at scale. This page estimates both in your browser, without sending your text anywhere.

What it does

• Estimates the token count of any text and shows how it splits between words, symbols and code, numbers and non-Latin text.
• Shows how much of an 8K, 32K, 128K, 200K and 1M context window the prompt fills.
• Works out the cost per request and per month from your reply length and request volume, for several models side by side.
• Lets you edit every price, add your own models and keep them in your browser for next time.
• Handles prompt caching and the batch discount, which can cut the bill a lot for repeated prompts.
• Opens text, Markdown, JSON, CSV and code files on your device.

Limits, honestly

• The count is an estimate, usually within 10 to 15 percent. Each provider’s tokenizer is different, and their vocabularies are too large to run in a web page. For an exact count use the provider’s own tokenizer or token-counting API.
• The prices were checked on 7 October 2026. Providers change them often and add models: check the pricing page and edit the numbers.
• It counts text only. Images, audio and files sent as attachments are priced differently by each provider.
• Hidden costs are not included: reasoning tokens some models spend before answering, tool calls and retries.
• Context windows listed are common sizes. A model’s real limit and its rules for output length are on its documentation page.