Converter
Tokens to words converter
Convert LLM token counts to approximate English word counts. Useful for sizing context windows, estimating output length, or planning document budgets.
Inverse of the ~1.33 words-per-token ratio. Assumes English prose.
How the conversion works
The inverse of the words-to-tokens ratio: 1 token ≈ 0.75 English words. This is the most useful direction when you're reading a model's context window in tokens and want to know how much actual content fits.
Example: Gemini 2.5 Pro's 2M-token context window holds roughly 1.5M words — about 15 novels. GPT-5's 400K window holds ~300K words, about 3 novels. Claude Opus 4.7's 200K window holds ~150K words, about 1.5 novels or 300 pages of single-spaced text.
Common token budgets in words
| 1,000 tokens | ~750 words (1-2 pages) |
| 8,000 tokens | ~6,000 words (a short paper) |
| 32,000 tokens | ~24,000 words (a short book) |
| 128,000 tokens | ~96,000 words (a novel) |
| 200,000 tokens (Claude) | ~150,000 words |
| 400,000 tokens (GPT-5) | ~300,000 words |
| 1,000,000 tokens | ~750,000 words (a trilogy) |
| 2,000,000 tokens (Gemini) | ~1.5M words |
Frequently asked
Is 0.75 words per token accurate?
For English prose, yes — within ±10% for most content. Code and non-English text will show a lower words-per-token ratio (more tokens per visible word).
How do I estimate output word length from a token budget?
Multiply by 0.75. If you ask a model to produce 1,000 output tokens, expect about 750 words of text — roughly 1.5 pages single-spaced.
Does the converter account for system prompts?
No. When you read a context window number (e.g. "200K tokens"), that budget has to hold your system prompt + user input + conversation history + tools + any reserved response space. Deduct 1,000-10,000 tokens from the headline number for realistic planning.
Why isn't the ratio exactly 1/1.33?
It is — 1/1.33 ≈ 0.752. We round to 0.75 for mental arithmetic; the difference is under 0.3% and well within the natural variance between documents.
Do different models give different word counts for the same output tokens?
Barely. All major tokenizers produce similar token-to-word ratios for English prose. A 1,000-token reply from GPT, Claude, or Gemini will all come out to roughly 700-800 words.
Ready to estimate a real prompt?
Paste your actual text into the estimator for exact token counts and dollar costs across every model.