Tokens to Words: How Many Words Is 100, 1,000, or 128K Tokens?

The short answer: 1 token ≈ 0.75 words

For ordinary English text, one token corresponds to about 0.75 words, or roughly 4 characters including spaces. Flip it around and one word costs about 1.3 tokens, and 100 tokens is approximately 75 words. These figures come from byte-pair-encoding (BPE) tokenizers, which compress frequent English patterns into single tokens and split rare ones.

Every serious token-to-words conversion you will find — from OpenAI's own guidance to calculator sites — is built on this family of ratios. The exact number for your text depends on the tokenizer and the writing style, which is why honest tools give a range rather than a single false-precision number.

Tokens to words conversion table

Here are the common token counts translated into words, characters, and pages. The typical column uses 0.75 words per token and 500 words per single-spaced page. Words ranges reflect the 0.65–0.85 spread you will see between casual prose and dense technical text.

• 100 tokens ≈ 75 words (65–85) ≈ 400 characters — a short paragraph
• 500 tokens ≈ 375 words (325–425) — a solid email
• 600 tokens ≈ 450 words (390–510) — a page of text
• 1,000 tokens ≈ 750 words (650–850) — a long blog section
• 2,000 tokens ≈ 1,500 words (1,300–1,700) — a full article
• 3,000 tokens ≈ 2,250 words (1,950–2,550) — a whitepaper section
• 4,096 tokens ≈ 3,072 words (2,662–3,482) — a short paper chapter
• 8,192 tokens ≈ 6,144 words (5,325–6,963) — a lengthy guide
• 16,384 tokens ≈ 12,288 words — a short ebook
• 65,536 tokens ≈ 49,152 words — a novella
• 131,072 tokens (128K) ≈ 98,304 words — a novel
• 200,000 tokens ≈ 150,000 words — a long novel
• 1,000,000 tokens ≈ 750,000 words — about 10 novels

Why the ratio is a range, not a law

Tokenizers reward frequency. Common words like the, and, and is compress to single tokens, while rare names, long numbers, URLs, and technical jargon shatter into multiple pieces. Casual prose with common vocabulary sits near 0.85 words per token; dense academic or legal text drifts toward 0.65.

Source code is the extreme case. Indentation, operators, identifiers, and symbols mean code can easily require 30–50% more tokens than English prose carrying the same visual length. If you are budgeting a coding agent, estimate code at roughly 3 characters per token instead of 4.

Non-English text changes the math

Languages written in Latin script with familiar words (Spanish, French, German) stay close to the English ratio. Languages that use different scripts or no spaces between words can cost several times more tokens per word — CJK text often lands near 1–1.5 characters per token. If your application is multilingual, measure each language separately instead of applying the 0.75 rule everywhere.

What token counts mean for context windows

Model context windows are quoted in tokens, which makes them hard to picture until you convert. A 200K-token window holds roughly 150,000 English words — a 300–500 page book. A 1M-token window holds about 750,000 words, or ten novels. Sudden context sizes sound abstract; book equivalents make planning concrete.

Remember the window is shared: instructions, conversation history, retrieved documents, tool results, and the model's answer all draw from the same budget. Reserve output space before deciding how much evidence fits.

Convert your own numbers

Static tables are useful for orientation, but real planning needs your actual text. The PromptCostLab tokens-to-words converter runs a real BPE tokenizer in your browser, counts exact tokens for OpenAI-family vocabularies, labels Claude/Gemini counts as close approximations, and shows words, characters, and pages side by side. Dedicated quick-answer pages exist for common counts like 600, 4,096, and 128K tokens.