Token Counter for AI Prompts
Paste your prompt or text and see an estimate of the token count with a realistic range. The figures are based on measurements with OpenAI tokenizers for the selected text language. Your text stays in the browser.
Source: OpenAI tiktoken (MIT) – “each token corresponds to about 4 bytes”; o200k_base / cl100k_base used for the measurement. Updated: .
How it is calculated
Tokens are the units a language model breaks text into – whole words, parts of words, punctuation, or spaces. Billing and limits (context window, maximum reply length) are measured in tokens, not characters.
How to count tokens
- Paste your prompt or text into the box.
- Choose the language of the text – German or French texts need more tokens than English ones.
- Read the estimated token count with its range, plus characters and words.
Example: With the rule of thumb of about 4 characters per token, an English text of 2,000 characters comes to roughly 2,000 / 4 = 500 tokens.
Rule of thumb: about 4 characters per token – in English
OpenAI says its tiktoken tokenizer averages about 4 bytes per token; Google states “about 4 characters” per token for Gemini. Both figures are for English text. Other languages need more tokens, because tokenizers are trained mostly on English text and words like “Stadtbibliothek” (German for “city library”) break into several pieces.
Where the numbers come from
We measured the OpenAI tokenizers o200k_base (GPT-4o and newer) and cl100k_base (GPT-4, GPT-3.5) on 39 to 56 plain-text samples per language (as of September 27, 2026). Average characters per token:
- English: 3.9 (range 3.3–4.5)
- German: 3.6 with o200k, 3.1 with cl100k (range 2.7–4.2)
- French: 3.7 and 3.3 (range 2.9–4.2)
- Spanish: 3.7 and 3.3 (range 2.9–4.2)
- Portuguese: 3.7 and 3.2 (range 2.7–4.2)
The estimate divides the character count by the average of both tokenizers; the range uses the 10th and 90th percentile of the individual texts. Program code, tables, lots of numbers, emoji, or languages with other scripts (Chinese, Arabic, …) can deviate significantly.
Why not an exact counter?
An exact tokenizer would need to load the full token table – for o200k_base that’s about 3.6 MB. That would be disproportionate for a fast page. Claude, Gemini, and Llama also use their own tokenizers; only the respective provider can give exact numbers (for example via a token-counting endpoint).
More text tools: count characters and words exactly with the word counter, strip extra spaces and line breaks with the text cleaner, and check how easy a text is to read with the readability checker.
Frequently asked questions
How many tokens are 1,000 words?
In English, roughly 1,300–1,400 tokens (measured average: 1.4 tokens per word). German needs more: depending on the tokenizer, about 1.8–2.1 tokens per word, so roughly 1,800–2,100 tokens.
Is the count accurate for ChatGPT?
No, it’s an estimate based on measurements with the ChatGPT tokenizers. In our test texts, 80–90% of the real values fell within the shown range; the midpoint was off by a median of 6–12%.
Does this also work for Claude or Gemini?
Roughly, yes, but every model has its own tokenizer. Google also states about 4 characters per token for Gemini in English. Exact numbers are only available through each provider’s own interface, such as an OpenAI, Claude, or Gemini token-counting endpoint.
Is my text uploaded anywhere?
No. Counting happens only in your browser; the text is never sent, never stored, and never written to the address bar.
Why are spaces and line breaks counted too?
Tokenizers also process whitespace: a space is usually attached to the following word, and multiple spaces or line breaks become their own tokens. That’s why the estimate is based on every character.
What is a token?
A token is the chunk of text an AI model such as ChatGPT processes – often part of a word. Common English words are usually one token, while long or rare words are split into several.
Why do other languages use more tokens than English?
Tokenizers are built mainly on English text, so words in German, French, Spanish or Portuguese are split into more pieces more often. That is why the counter asks for the language of your text.
Sources and legal basis
- OpenAI tiktoken (MIT) – “each token corresponds to about 4 bytes”; o200k_base / cl100k_base used for the measurement
- Google AI for Developers – Understand and count tokens (“about 4 characters”)
- Anthropic – Token counting (exact counts via the Claude API)
As of: · Editorial policy
Related tools
- Word Counter and Character CounterCharacter counter: characters with and without spaces, words, sentences, reading and speaking time, top words – live, with limits for X, Instagram and SMS.
- Password GeneratorGenerate secure random passwords and memorable passphrases with entropy shown in bits — calculated right in your browser, nothing sent or stored.
- Password Strength CheckerHow secure is my password? Check strength, entropy in bits, and how long it would take to crack — locally in your browser, never sent or stored.
- Text to Speech: read text aloudText to speech online: paste text or load a file, pick a voice, speed and pitch, follow along word by word and skip by sentence. Free, no sign-up needed.
- Morse Code TranslatorMorse code translator: text to Morse and back, live. Play it as sound or light, download a WAV, tap your own code on a Morse key. With full ITU chart.
- Flesch Reading Ease CalculatorPaste your text to get the Flesch Reading Ease score, the score band, words per sentence and syllables per word – plus the German version by Amstad.
- Handwriting and name tracing worksheet generatorMake free tracing worksheets: letters, words or names in print or cursive on handwriting lines (1 in, 5/8 in, 1/2 in). Printable PDF for preschool to grade 3.
- Reading Time CalculatorReading time calculator: paste a text and see at once how long it takes to read and to read aloud – with word count, page count and an adjustable reading speed.
- Remove duplicate linesRemove duplicate lines from lists and text, show only unique or only repeated lines, with counts and sorting. Runs in your browser, nothing is uploaded.