All Tools Verse1,000 tokens is about 750 English words. The rule of thumb comes from OpenAI’s help center article “What are tokens and how to count them?”: one token is about three quarters of a word, or about four characters. The real number depends on the tokenizer, the language and the kind of text, so treat 750 as an estimate, not a count.
In this guide
Tokens to words table
This table uses 0.75 words per token, 4 characters per token, and 500 words for a single-spaced page.
| Tokens | Words (about) | Characters (about) | Single-spaced pages |
|---|---|---|---|
| 100 | 75 | 400 | 0.15 |
| 500 | 375 | 2,000 | 0.75 |
| 1,000 | 750 | 4,000 | 1.5 |
| 2,000 | 1,500 | 8,000 | 3 |
| 4,000 | 3,000 | 16,000 | 6 |
| 8,000 | 6,000 | 32,000 | 12 |
| 32,000 | 24,000 | 128,000 | 48 |
| 128,000 | 96,000 | 512,000 | 192 |
| 200,000 | 150,000 | 800,000 | 300 |
| 1,000,000 | 750,000 | 4,000,000 | 1,500 |
The tokens to words converter uses the same ratios. Type any number of tokens, words, characters or pages and it fills in the rest, and shows whether the text fits common context window sizes.
What we measured
We ran sample texts through tiktoken, the open-source tokenizer library from OpenAI, with two of its encodings: cl100k_base (used by GPT-4 and GPT-3.5 Turbo) and o200k_base (used by GPT-4o). Words were counted by splitting on spaces.
| Text | Words | Tokens (cl100k) | Tokens (o200k) | Words per token (o200k) |
|---|---|---|---|---|
| Plain English, 5 paragraphs | 269 | 310 | 307 | 0.88 |
| Technical English, 1 sentence | 16 | 31 | 30 | 0.53 |
| Spanish, 2 sentences | 36 | 51 | 43 | 0.84 |
| German, 2 sentences | 30 | 55 | 41 | 0.73 |
| Greek, 2 sentences | 32 | 161 | 70 | 0.46 |
| Python code, 4 lines | 19 | 41 | 41 | 0.46 |
| Order number, date and tracking code | 11 | 41 | 41 | 0.27 |
Three things stand out. Simple everyday English beat the rule of thumb: 1,000 tokens held about 876 words. Technical English, code and codes full of digits fell well below it. And the same two sentences about a train timetable cost 33 tokens in English, 43 in Spanish and 70 in Greek with the newer encoding. With the older cl100k encoding, the Greek version took 161 tokens.
So 750 words per 1,000 tokens is a cautious middle value for English. It is a safe planning number, but it is wrong in both directions for some texts.
Why the ratio changes
- Common words are one token. Words like “the”, “train” and “office” are usually single tokens, often with the space in front of them included.
- Rare and long words split. With o200k_base, “Asynchronous” becomes “As” plus “ynchronous”, and “acknowledgements” also takes two tokens.
- Numbers and codes split into short chunks. The 18-character tracking code 1Z999AA10123456784 takes 9 tokens, one for every two characters.
- Tokenizers are trained mostly on English. Other languages, and especially non-Latin scripts, need more tokens for the same meaning. Newer encodings with larger vocabularies narrow the gap.
- Each model family has its own tokenizer. Our measurements are for OpenAI encodings. Claude, Gemini and Llama models use different tokenizers and give different counts for the same text.
Words to tokens
To go the other way, divide the word count by 0.75, which is the same as multiplying by about 1.33.
| Words | Tokens (about) |
|---|---|
| 300 (a short email) | 400 |
| 1,000 | 1,333 |
| 1,500 | 2,000 |
| 2,500 | 3,333 |
| 80,000 (a novel) | 106,667 |
To get the word count of your own text first, paste it into the word counter. If you prefer the 4-characters rule, the character counter gives you the character count to divide by 4. For an exact count with a specific tokenizer, use the AI token counter.
Common mistakes
- Using the estimate for billing. API costs are charged on the real token count. For a budget, run a sample of your actual text through a tokenizer.
- Counting only your prompt. The context window holds the system prompt, the conversation so far and the reply. A 4,000-word reply needs about 5,333 tokens of room on top of what you send.
- Assuming English ratios for other languages. In our test, Greek used more than twice the tokens of English for the same sentences.
- Mixing page sizes. A double-spaced page holds about 250 words, half a single-spaced page. 1,000 tokens is about 1.5 single-spaced pages or 3 double-spaced pages.
More text guides and tools: free text tools.
Frequently asked questions
How many tokens is 1,000 words?
About 1,333 tokens for English, using 0.75 words per token. Plain prose often needs fewer, and technical text more.
How many tokens is one page?
A single-spaced page of about 500 words is roughly 667 tokens. A double-spaced page of about 250 words is roughly 333 tokens.
Do spaces and punctuation count as tokens?
Spaces are usually merged into the word that follows, so they rarely cost a token on their own. Punctuation marks are often separate tokens. Line breaks and indentation in code can add tokens too.
Is the ratio the same for Claude or Gemini?
No. Each provider uses its own tokenizer, so the same text gives a different count. The 0.75 rule is a fair first guess for English, but check with the provider’s own token counter when the number matters.