BroadRouter

Tokens to words converter

Updated

In short: 1,000 English words is about 1,290 tokens, and 1,000 tokens is about 775 words (averaged over five tokenizers, about 4.6 characters per token). Other content needs more: 1,000 words of Modern Standard Arabic is about 2,080 tokens and 1,000 words of Python code about 2,126. Convert tokens, words, characters or pages for your content type.

Free, no sign-up, runs in your browser: nothing you type is sent anywhere.

Convert
Measure my text (real tokenizer)

Runs in your browser. The tokenizer downloads once from Hugging Face; nothing you paste is sent.

TokenizerTokensWordsCharactersPages

Report a wrong price or number (email to info@broadnet.ai, with this page's link and your inputs filled in).

How it works

  1. Type a number and choose what it is: tokens, words, characters or pages.
  2. Pick the content type (English prose, Arabic, code, JSON, Chinese and more) and the model's tokenizer.
  3. Read the conversion into the other three units, and the same conversion for every tokenizer in the table.
  4. Optional: paste your own text under "Measure my text" to get a ratio from your writing with the real tokenizer. Copy the link to share.

How many tokens is 1,000 words?

About 1,290 tokens for English prose, averaged over five real tokenizers (DeepSeek, GLM, Qwen, Gemma and GPT o200k). The answer depends on the content: Arabic, code and JSON need far more tokens per word.

ContentTokens per 1,000 words (average)Range across tokenizers
English prose1,2901,269 to 1,320
Arabic (Modern Standard)2,0801,830 to 2,439
Gulf Arabic2,2682,049 to 2,507
Egyptian Arabic2,1031,897 to 2,437
Levantine Arabic (North)2,2121,972 to 2,562
Arabizi (Arabic in Latin letters)2,2472,098 to 2,336
French1,5341,436 to 1,635
Spanish1,4291,347 to 1,548
Hindi2,5301,380 to 4,547
Urdu2,0061,488 to 2,683
Chinese (Simplified)1,2281,109 to 1,478
Code: Python2,1262,004 to 2,339
Code: JavaScript1,8521,755 to 1,986
Code: SQL2,2482,098 to 2,448
JSON4,3593,888 to 4,944
Russian (small v1 sample)1,8111,711 to 1,938

Measured 2026-11-15 on the open-licensed texts of our dataset (Russian: an older, smaller sample). The full data, with every count and the sample texts, is in the tokenizer efficiency dataset (CSV and JSON).

Measured ratios by content type

Rules of thumb such as "a token is three quarters of a word" only hold for English, and only roughly. We measured each content type with each real tokenizer. Words per 1,000 tokens, higher means more text per token:

ContentSample wordsDeepSeekGLMQwenGemmaGPTChars per token (avg)
English prose11837857757587707884.61
Arabic (Modern Standard)390374504105474935302.84
Gulf Arabic2034323994524434882.46
Egyptian Arabic385474444105275015162.77
Levantine Arabic (North)350844223905074684952.72
Arabizi (Arabic in Latin letters)2144394454284394772.41
French507456126246676696964.03
Spanish520226466677177427384.20
Hindi514863412204657246112.01
Urdu548754343735106726252.33
Chinese (Simplified)456498908799027746761.56
Code: Python2804754994644274943.72
Code: JavaScript4905575595165045703.65
Code: SQL3574484774264084733.82
JSON17062492432072022572.84
Russian (small v1 sample)1285165495845575593.83

Method. Parallel text: FLORES-200 dev and devtest (2,009 aligned sentences per language, CC BY-SA 4.0, tarball pinned by SHA-256) for Modern Standard, Egyptian, North Levantine, Najdi and Mesopotamian Arabic, Hindi, Urdu, Chinese, French and Spanish against English, each sentence counted on its own; Tatoeba sentence pairs (CC BY 2.0 FR) for Gulf Arabic and Arabizi. Ratio = pooled tokens ÷ pooled tokens of the English side, with a 95% interval from a seeded bootstrap over sentences (10,000 resamples). Other texts: Wikipedia lead sections (CC BY-SA 4.0), Project Gutenberg (public domain) and Wikidata JSON (CC0), stored with source URL, licence and revision. Tokens: the official tokenizer file at a pinned Hugging Face commit, encoded with @huggingface/tokenizers with no special tokens; o200k_base via gpt-tokenizer; Kimi's tiktoken rank file with a tiktoken-compatible counter. Gated Meta and Google files were read from ungated mirrors proven byte-identical to the official file. Every new count matched the Rust tokenizers library or Python tiktoken. Words: word-like segments from Intl.Segmenter. Characters: Unicode code points. A page is 500 words. Measured 2026-11-15. Rerun with node scripts/dataset-tokenizer-efficiency.mjs.

The formulas

words      = tokens × (sample words ÷ sample tokens)
characters = tokens × (sample characters ÷ sample tokens)
pages      = words ÷ 500
tokens     = words  × (sample tokens ÷ sample words)

Worked example: with the DeepSeek V4.1 Flash tokenizer, the English sample is 1183 words and 1507 tokens. So 1,000 tokens ≈ 1,000 × 1183 ÷ 1507 = 785 words, or 1.6 pages.

Words are counted the way a word processor would for space-separated languages. For Chinese, words come from the browser's dictionary-based word segmenter, so "words" there are multi-character words rather than single characters. For a precise count of a specific text, use the token counter; to see whether a document fits a model, use the context window calculator.

Questions

How many words is 1,000 tokens?

About 775 words of English prose with today's tokenizers (our measurement, averaged over five tokenizers). The common rule of thumb, 750 words per 1,000 tokens, is on the safe side for plain English prose with modern tokenizers. Code, JSON and most non-Latin scripts give far fewer words per token.

How many tokens is 1,000 words?

About 1,290 tokens for English prose, 2,080 for Modern Standard Arabic and 2,126 for Python code, averaged over the five tokenizers we measured.

How many characters is a token?

About 4.6 characters per token for English prose, including spaces. Arabic is about 2.8 and Chinese about 1.6, since one Chinese character carries a whole syllable.

How many tokens is a page?

At 500 words per page (single-spaced), a page of English prose is about 645 tokens. A page of code or JSON is much more because symbols and indentation are tokens too.

Why does the ratio depend on the model?

Each model family has its own tokenizer. Larger, more multilingual vocabularies pack non-English text into fewer tokens. Pick the tokenizer above, or use the token counter for an exact count of your text.

How accurate is this converter?

The ratios are measured, not guessed, but they come from sample texts and your text will differ. Even within English, short everyday sentences needed 2% more tokens per word than long-form prose in our dataset. Use "Measure my text" for a ratio from your own writing, or the token counter for an exact count.

Go deeper

Read as MarkdownAsk ChatGPTAsk ClaudeAsk Perplexity