Every major LLM API bills by tokens, not words or characters, but most people think in words when drafting a prompt. This calculator converts between the two using the standard English-text rule of thumb (roughly 750 words per 1,000 tokens) so you can estimate token usage before you write a single line of API code.
How it works
The conversion uses the widely-published approximation that 1,000 tokens is about 750 words of typical English text. Equivalently, 1 token is roughly 0.75 words, or about 4 characters.
This is a heuristic, not an exact count. Real tokenizers (like OpenAI's tiktoken or Anthropic's tokenizer) split text into sub-word pieces based on the model's training vocabulary, so punctuation, numbers, code, non-English text, and unusual formatting can all shift the real token count noticeably above or below this estimate.
For a quick planning estimate (sizing a prompt, checking it fits a context window, or ballparking cost before using the AI API Cost Calculator), this approximation is normally close enough. For exact billing-critical counts, use the specific model provider's official tokenizer.
A worked example
750 words of typical English prose converts to an estimated 1,000 tokens and roughly 4,000 characters, a useful ballpark before checking a prompt against a model's exact context window limit.
Questions people ask
Why doesn't 1 token equal 1 word?
Tokenizers split text into sub-word pieces, not whole words. Common words are often a single token, while longer or less common words split into multiple tokens. On average this works out to roughly 0.75 words per token for typical English text, which is where the ~750-words-per-1,000-tokens estimate comes from.
Does this work for code or non-English text?
Less accurately. Code tends to use more tokens per character than prose because of symbols, indentation, and identifiers; non-English languages (especially non-Latin scripts) often tokenize less efficiently than English, sometimes needing 2-3x more tokens for the same apparent amount of text.
How do I get an exact token count instead of an estimate?
Use the specific provider's own tokenizer. OpenAI publishes tiktoken, and Anthropic and Google provide their own token-counting endpoints and libraries. This calculator is for quick planning, not for a number you'd want to bill a client against.
How does this relate to the AI API Cost Calculator?
This tool estimates how many tokens your text will use; the AI API Cost Calculator turns a token count into a dollar cost for a specific model. Use this one first if you only know your prompt's word count, then plug the resulting token estimate into the cost calculator.