The rule of thumb
In English, 1,000 tokens are about 750 words on GPT-5.6 (737 with our averages) and around 510 on Claude 5.x models, whose newer tokenizer splits text into smaller pieces.
words = tokens × characters per token ÷ characters per word
We use 5.7 characters per word in English (5.9 in Spanish and 6.0 in Portuguese, counting the space) and 500 words per page.
What it's for
- Estimating how many documents fit in a prompt before you build RAG.
- Turning output limits ("up to 8,000 tokens") into text length.
- Pricing a run over a library of books or PDFs.