Tokens to Words Converter
Turn a token count into words and pages, or words into tokens, for GPT, Claude and Gemini in 41 languages.
Need an exact count for a real text? Paste it into the token counter.
Quick reference (everyday English)
A page is about 500 words, so 1,000 tokens on GPT is roughly 1¾ pages and a 128,000-token context window holds about 220 pages.
| Tokens | GPT | Claude Opus / Sonnet / Haiku 5.5 | Gemini |
|---|---|---|---|
| 1,000 tokens | 870 words | 630 words | 870 words |
| 10,000 tokens | 8,696 words | 6,301 words | 8,696 words |
| 128,000 tokens | 111,304 words | 80,655 words | 111,304 words |
| 1 million tokens | 869,565 words | 630,120 words | 869,565 words |
Where the numbers come from
We counted real English text on GPT's current tokenizer (o200k): ordinary prose came out at 1.12–1.15 tokens per word, technical documentation at 1.33, about 4.4–4.9 characters per token. For Claude and Gemini we use our own measurement of the same prompt in 41 languages, counted with each provider's token-counting API: in English, Claude Opus, Sonnet and Haiku 5.5 used 38% more tokens than GPT, Claude Haiku 4.5 3% more, and Gemini the same. The gap changes by language (Korean +82% on Claude 5.5, Japanese +16%), and the converter uses the figure for the language you pick. The full method is in How many tokens is a word?
Other languages use the ratios from our 41-language measurement: the same prompt translated into each language and counted on GPT. Words are a poor unit across languages (Chinese, Japanese and Thai do not separate words with spaces), so for them the converter shows how much English-equivalent content fits in your tokens.
Related
- AI token counter — exact GPT counts and API cost for any text
- What is a token?
- Context windows explained
- Same prompt, 41 languages: token cost compared
How many words is 1,000 tokens?
In everyday English on GPT models, 1,000 tokens is about 870 words, or roughly 1¾ pages. Technical writing is closer to 750 words. On Claude Opus, Sonnet and Haiku 5.5 the same 1,000 tokens hold only about 630 words, because in our test Claude used 38% more tokens than GPT for the same English text.
How many tokens is 1,000 words?
About 1,150 tokens on GPT for ordinary English prose, and about 1,330 for technical text. The same 1,000 words are about 1,590 tokens on Claude Opus, Sonnet and Haiku 5.5, and about the same as GPT on Gemini.
Is "1 token = ¾ of a word" still right?
It is a safe, slightly high estimate. That rule of thumb comes from OpenAI's older tokenizers. On current GPT models we measured 1.12 to 1.33 tokens per English word, so ordinary writing usually uses about 15% fewer tokens than the rule predicts.
Why do other languages need more tokens?
Tokenizers are trained mostly on English, so other scripts get split into smaller pieces. For the same meaning, Korean needs about 44% more tokens than English and Japanese about 79% more, while Chinese is close to English. Pick a language above to see the effect.
Is this exact?
No, it is an estimate from average ratios. For the exact count of a specific text, paste it into the token counter, which runs the real GPT tokenizer in your browser.