Tokens per word
For English writing, a useful planning range is about 1.3 to 1.5 tokens per word. It is close enough for early sizing, not for billing.
Token and word conversion
Tokens to Words Calculator
Use this page when you have a token budget but need a plain-language sense of size: a short brief, a long article, a document batch, or even 1 million tokens expressed as words.
Conversion inputs
Text type
Estimated words
670,000 - 770,000
1,000,000 tokens using english prose
Estimated tokens
975,000 - 1,125,000
750,000 words in English prose
Common token budgets to words
Rough English prose estimates for planning. Paste final text into the token calculator when precision matters.
| Token budget | Estimated words | Typical planning use |
|---|---|---|
| 1,000 tokens | 670 - 770 words | Short prompts, emails, or small excerpts |
| 10,000 tokens | 6,700 - 7,700 words | Long prompts, notes, or compact documents |
| 100,000 tokens | 67,000 - 77,000 words | Large document chunks or research packets |
| 1,000,000 tokens | 670,000 - 770,000 words | Very large contexts or batch-planning estimates |
1 million tokens to words
One million tokens is a very large text budget. In prose it can mean hundreds of thousands of words; in code, the word estimate is much less tidy.
Exact prompts
Ratios are useful when you are sketching. For context limits, API checks, or cost estimates, measure the actual prompt.
Tokens to words FAQ
How many words is 1 million tokens?
For plain English prose, 1 million tokens is often somewhere around 670,000 to 770,000 words. Code, JSON, dense formatting, and non-English text can land outside that range.
How many tokens are in one word?
In English prose, one word often works out to about 1.3 to 1.5 tokens. Punctuation, whitespace, language, and the tokenizer itself can all move the ratio.
Should I use words or tokens for LLM limits?
Use tokens for anything tied to API limits, context windows, or billing. Word count is fine for early planning, but final prompts should be checked as tokens.
Why do code and JSON have different token-to-word ratios?
Code and JSON are full of braces, keys, punctuation, indentation, symbols, and repeated short strings. Tokenizers do not treat that like ordinary prose.