How many tokens is 1,000 words? A practical guide to tokens

1,000 English words come to roughly 1,250 to 1,670 tokens, and about 1,430 is a good single figure to plan with. That range comes from Google’s Gemini documentation, which says 100 tokens are about 60 to 80 English words and a token is about four characters. The exact number depends on the model’s tokenizer and on what you write: plain prose sits at the low end, code, numbers and most non-English text at the high end.

What a token is

A language model does not read letters or words. It reads tokens: pieces of text from a fixed vocabulary that the model’s tokenizer learned before training. Common English words are usually one token each, often with the space in front of them included. Longer or rarer words split into two or more pieces. Punctuation, digits and line breaks are tokens too.

OpenAI’s o200k_base tokenizer, the one used by GPT-4o, GPT-4.1, GPT-5 and the o-series, cuts the sentence “Hello world! Tokenization splits text into pieces.” into 10 tokens: Hello, world, !, Token, ization, splits, text, into, pieces, .. Seven words became ten tokens, which is exactly the 70-words-per-100-tokens middle of Google’s range.

Common lengths in tokens

Text	Words	Tokens (60 to 80 words per 100 tokens)	Planning figure
A short email	120	150 to 200	about 170
A 1,000-word blog post	1,000	1,250 to 1,670	about 1,430
A 10-page report (500 words a page)	5,000	6,250 to 8,330	about 7,140
A 50-page contract	25,000	31,250 to 41,670	about 35,700
A 300-page book	150,000	187,500 to 250,000	about 214,000

Going the other way, a 4,000-token answer is roughly 2,400 to 3,200 words, and a 128,000-token context window holds roughly 77,000 to 102,000 words of English.

Why your text may need more tokens

Code

Code is full of symbols, indentation and names that are not ordinary words. Brackets, operators and runs of spaces each take tokens, so a page of code usually needs more tokens than a page of prose with the same number of characters.

Languages other than English

Tokenizers learn their vocabulary from training text, and much of it is English. Words in other languages, and especially in non-Latin scripts, tend to be split into more and smaller pieces. The same message can cost noticeably more tokens in Arabic, Hindi or Chinese than in English. The only reliable way to know is to measure a sample of your own text.

Numbers, tables and formatting

Long numbers are split into several tokens, and so are dates, IDs and URLs. Markdown tables, JSON and HTML add structure characters that all count.

Do all models count the same way?

No. Each provider uses its own tokenizer, so the same text gives slightly different counts on GPT, Claude, Gemini, DeepSeek or Mistral models. OpenAI publishes its tokenizers, which is why GPT counts can be exact; OpenAI has not published the tokenizer of its GPT-6 models. Anthropic and Google offer token-counting endpoints in their APIs for exact counts on their models. For planning, a count from one modern tokenizer is close enough; for billing questions near a limit, use the provider’s own count.

Why tokens matter

• Price. API providers charge per million tokens, with separate prices for input and output. Output is the more expensive side on every major model.
• Limits. Every model has a context window, the number of tokens it can hold at once, covering your prompt, any documents, the chat so far and the answer.
• Speed. Models write one token at a time, so a longer answer takes longer to arrive.

Count your own text

Two free tools on this site do the counting for you, in your browser:

• The words to tokens converter turns words, pages or characters into tokens and back, and can measure the exact ratio of a sample of your own writing.
• The token counter counts pasted text or a text file with OpenAI’s o200k tokenizer, shows how it splits, and prices it on each model.

Sources

• Google, Understand and count tokens (Gemini API documentation), read October 26, 2026.
• OpenAI, tiktoken model-to-encoding map, version 0.14.0.

Free toolHow long is a 4,000-token answer?Convert