Token Counter: Estimate Tokens for ChatGPT, Claude and Gemini
Chat shortcut: /token
Paste a text into the token counter and see at once roughly how many tokens it uses, how many words and characters it has and what it costs as a question in the chat.
No AI No Sign-Up
Estimated tokens
0
0 Words · 0 Characters
ChatX tokens if this text is sent as a question
The estimate counts real model tokens. ChatX charges ChatX tokens: the same amount converted with a factor based on the model’s price. Cheap models need few ChatX tokens, expensive ones more. The answer and the previous conversation are added. Overview of models and costs
Why count tokens and how ChatX bills
Language models do not read and write in words but in tokens: small building blocks such as a short word, part of a word, a punctuation mark or a space. Every model splits text a little differently, so the same paragraph yields different numbers per model. The counter averages a character-based and a word-based estimate; it uses no provider's tokenizer and therefore gives an order of magnitude, not an exact value.
On ChatX you pay per answer based on actual usage: the input (your message plus the history sent along, which ChatX limits by message and token budget) and the model's output, and with thinking models also the thinking phase. The usage is converted into ChatX tokens according to the model's prices; input, output and cached input each have their own price. The list next to the field shows per model what the input alone would cost as a question; answer and history come on top.
This is useful above all with long texts: to check whether a message fits your character limit, to split a document sensibly, or to estimate the order of magnitude of a large job before sending.
How it works
- 1
Paste the text
Copy your text into the field; the numbers update as you type.
- 2
Read the values
Check the estimated tokens, words and characters; below, per model, you see roughly how many ChatX tokens the input costs as a question.
- 3
Choose a model
Compare the models in the list and decide whether to shorten or pick a cheaper model; prices for answers and cache are in the overview of models and costs.
Typical uses
Check the character limit
See before sending whether a long message fits your limit or should be shortened.
Split documents
Divide a long text into sections the model can process in one pass.
Estimate costs
Before a large job, estimate how much credit the input ties up before the answer is added.
Tighten prompts
Cut system instructions and templates down to what is needed and see the effect in tokens immediately.
Good to know
- Runs entirely in the browser: the text is not transmitted anywhere and not stored.
- Free and without sign-up, with no limit on text length in the counter itself.
- Shows estimated tokens, words and characters from the average of a character-based and a word-based estimate; the estimate does not replace a provider's tokenizer, deviations depend on text and model.
- Languages other than English, German for example, usually need more tokens per text because accented characters and long compound words split into several building blocks; code, tables and emojis can also need more tokens.
- Before sending, ChatX only estimates roughly; billing is based on the usage the provider reports after the answer. Answer and history sent along count too; thinking phases count as output.
- The per-model list shows only the input share; it does not account for the answer price, cache benefits or free models. The complete prices are in the overview of models and costs.
Frequently asked questions
What is a token?
Language models split text into small building blocks, the tokens. A German word corresponds on average to about 1.5 tokens, an English word to slightly fewer.
How accurate is the estimate?
The counter averages a character-based and a word-based estimate without a provider's tokenizer; the real value can deviate from it depending on model and text. On ChatX, billing is based on the usage the provider reports; if that figure is missing after an interruption, it is estimated.
Why do other languages need more tokens than English?
The models' tokenizers are trained heavily on English. Words with accents or umlauts and long compounds, as in German, are split into more parts, which is why the same content in other languages usually costs more tokens.
Does the model's answer count too?
Yes. You pay for input and output; with most models the output is more expensive than the input. With models that have a thinking mode, the thinking phase also counts as output. The counter and its model list show only the input.
What is the difference between model tokens and ChatX tokens?
Model tokens are the building blocks the provider counts. ChatX tokens are derived from them via the model's prices: input, output and cached input are converted according to their price. So a cheap model costs few ChatX tokens for the same text, an expensive one more.
How do I save tokens in the chat?
Start a new chat for a new topic so the old history is not sent along, shorten introductions and repetitions, and switch on thinking mode only when the task needs it.
Is my text stored?
No. The counter works in the browser; the text never leaves your device.