Convert tokens to words—and words to tokens.
Translate an abstract context-window number into a readable workload range. Choose a content profile because prose, code, JSON, and multilingual text do not share one universal conversion ratio.
Natural-language paragraphs with ordinary punctuation and spacing.
750,000
estimated words
1,000,000 tokens × 0.75 words/token
Have the actual content? Tokenize it instead of estimating.
Open the exact content calculator →A planning interval, not fabricated precision.
The converter models words per token rather than claiming that every token equals a fixed fraction of a word. English prose starts from a 0.75 words-per-token center with a 0.70–0.80 interval. More syntactically dense profiles use lower centers and wider intervals.
In the reverse direction, the same profile is inverted: estimated tokens = words ÷ words per token. The calculation is deterministic, but its applicability depends on how closely the unknown content matches the chosen profile.
Review the token measurement methodology →Use the content when you have the content.
Frequently asked questions
How many words are 1 million tokens?+
For ordinary English prose, one million tokens is commonly about 750,000 words, with a planning interval of roughly 700,000 to 800,000 words. Code, structured data, technical notation, and multilingual text can differ substantially.
Is converting tokens to words exact?+
No. A token count alone does not contain the original words, language, whitespace, or punctuation. Token-to-word conversion is a workload estimate; exact tokenization requires the actual content and a model tokenizer.
Why does source code produce fewer word-equivalents per token?+
Identifiers, operators, punctuation, indentation, escaped strings, and compact syntax consume tokens without mapping cleanly to natural-language words.
Can I calculate exact tokens from my document?+
Yes. Use the main TokenCalculator.dev calculator with the actual text, PDF, DOCX, code, data file, or supported image. Processing runs locally in the browser.