How many words is 1,000 tokens?
The short answer, the conversion tables both ways, and the cases where the usual rule of thumb stops holding.
1,000 tokens is about 750 words of ordinary English, or roughly 4,000 characters. Going the other way, 1,000 words is about 1,333 tokens.
A token is the unit an AI model reads text in. It is usually a short word or a fragment of a longer one, which is why the count never lines up exactly with words. For plain English the ratio is stable enough to plan around: one token averages about four characters, and about three quarters of a word.
Tokens to words
| Tokens | Words | Characters |
|---|---|---|
| 100 | 75 | 400 |
| 500 | 375 | 2,000 |
| 1,000 | 750 | 4,000 |
| 2,000 | 1,500 | 8,000 |
| 4,000 | 3,000 | 16,000 |
| 8,000 | 6,000 | 32,000 |
| 16,000 | 12,000 | 64,000 |
| 32,000 | 24,000 | 128,000 |
| 128,000 | 96,000 | 512,000 |
Words to tokens
| Words | Tokens | Characters |
|---|---|---|
| 100 | 133 | 533 |
| 250 | 333 | 1,333 |
| 500 | 667 | 2,667 |
| 750 | 1,000 | 4,000 |
| 1,000 | 1,333 | 5,333 |
| 2,000 | 2,667 | 10,667 |
| 5,000 | 6,667 | 26,667 |
| 10,000 | 13,333 | 53,333 |
How many tokens is a page?
A double spaced page runs about 250 words, so roughly 333 tokens. A single spaced page runs about 500 words, so roughly 667 tokens. A 10 page report of 5,000 words lands near 6,700 tokens, which still fits comfortably inside any current context window.
When the rule of thumb breaks
Other languages
English is the cheapest language per unit of meaning, because tokenizer vocabularies are built mostly from English text. Arabic, Chinese, Japanese, Korean, Thai, and other non Latin scripts split into much shorter pieces, so the same paragraph can cost two to three times as many tokens. If you are budgeting a prompt in one of those languages, do not trust an English ratio.
Code, numbers, and punctuation
Source code is dense with symbols, indentation, and identifiers that no vocabulary covers whole, so it commonly runs closer to 3 characters per token. Long digit strings are worse still: numbers are split into small chunks, so a table of figures can cost far more than the same space of prose.
Different models, different tokenizers
Every model family ships its own tokenizer, so GPT, Claude, Gemini, and Llama will each return a slightly different number for the same sentence. For English prose the spread is small, usually a few percent. For code and non Latin scripts it widens. Any counter that runs outside the model itself, including this one, is giving you an estimate rather than a billing figure.
Questions
How many words is 1,000 tokens?
About 750 words of plain English, or about 4,000 characters.
How many tokens is 1,000 words?
About 1,333 tokens. Multiply your word count by roughly 1.33, and by more if the text is heavy with numbers or code.
How many tokens is one page of text?
Roughly 333 tokens for a double spaced page, roughly 667 for a single spaced one.
Why does the token count differ between models?
Each model family uses its own tokenizer vocabulary, so the same text splits differently. The gap is small for English prose and much wider for code and non Latin scripts.
Do non English languages use more tokens?
Usually yes, often two to three times more for the same meaning.
Count the real number
Rules of thumb are for planning. When you need the count for the text actually in front of you, paste it into the counter: characters, words, lines, paragraphs, tokens, and speaking time all update as you type, in your browser, with nothing uploaded.
iLostCount is a free character, word, and token counter. No ads, no signup, no tracking beyond cookieless page view counts.
Back to the counter