Free tool · no sign-up

Tokens to pages converter: how many pages is your prompt?

A 500-word page of English prose is about 665 tokens at 1.33 tokens per word. A 200K-token window holds roughly 300 pages before you reserve output; after an 8K output reserve, 2K system prompt and 2K margin, usable input is about 282 pages. Convert a value below or use the reference table.

Tokens199,500
Words150,000
Characters798,000
Pages300
8K window
100%
32K window
100%
128K window
100%
200K window
100%
1M window
20%

Estimates use 1.33 tokens per word and 4 characters per token. Context bars show the raw window; the usable input is lower after output and prompt overhead.

Tokens, words, characters and pages

DocumentWordsEstimated tokensCharacters200K window1M window
1 page5006652,6600.3%0.1%
10 pages5,0006,65026,6003.3%0.7%
50 pages25,00033,250133,00016.6%3.3%
150 pages75,00099,750399,00049.9%10.0%
300 pages150,000199,500798,00099.8%20.0%
500 pages250,000332,5001,330,000166.3%33.3%

Decimal percentages of the raw context window. A 300-page report at 199,500 tokens looks like it fits a 200K window, but does not fit after the output reserve. See the context-window calculator for the usable-input check.

How many tokens is a page?

For English prose, the working approximation is one token per four characters, or about 0.75 words per token. That gives 1.33 tokens per word and roughly 665 tokens for a 500-word page.

Three caveats matter:

  • Prose varies. Vocabulary, punctuation and formatting can move the estimate by ±10–15%.
  • Code is denser. JSON, YAML and source code often run closer to three characters per token, so a 100 KB JSON file is nearer 33,000 tokens than 25,000.
  • Non-Latin text differs. Chinese, Japanese and Korean can exceed one token per character on some tokenizers. Do not apply the English ratio to a multilingual corpus.

Why the page count is not the context fit

The context window is the total budget for one request: system prompt, tool schemas, retrieved documents, conversation history, the user message and the generated answer. The page count only describes the document. To check fit:

usable_input = context − output_reserve − system_prompt − tool_schemas − safety_margin

On a 200K model with an 8K output reserve, 2K system prompt and 2K margin, usable input is 188,000 tokens, or about 282 pages. A 300-page report is 11,500 tokens over and needs chunking or summarisation.

When to use a real tokenizer

Use this converter for planning. Use a tokenizer when the result drives a hard limit or a bill:

  • OpenAI models: run the matching tiktoken encoding locally for an exact count.
  • Open-weight models: use the Hugging Face tokenizer that ships with the model.
  • Anthropic and Google: use the provider's token-count endpoint where available.

Encodings and endpoints change; confirm the mapping against the provider's current documentation. This page labels its output as an estimate for that reason.

Book pages vs manuscript pages vs report pages

“Page” is not a standard unit. A publisher may set a page at 250–300 words; a double-spaced manuscript page is about 250 words; this converter uses 500 words for a report page. That difference changes the answer by 2×.

If someone says “a 300-page book,” ask which definition they mean. At 300 words per page, 300 pages is 90,000 words, or about 119,700 tokens — not the 199,500 tokens this page assigns to 300 report pages. Use the same definition for the document and the context check.

How to reduce a document's token count

  • Summarise before sending. A 20-page section can become a 300-word summary plus the source links.
  • Retrieve, don't dump. Send the 5–10 chunks that answer the question, not the whole corpus.
  • Trim tool schemas. Nested JSON definitions can consume thousands of tokens before the user says anything.
  • Shorten history. Keep the last few turns plus a rolling summary; do not resend the entire transcript.
  • Lower the output reserve. Reserve the p95 completion length plus headroom, not the model's maximum output.
  • Cache the stable prefix. Caching cuts cost, but usually not the token count against the context window.

Code, tables and non-Latin text

The 4-characters-per-token rule is for English prose. Code often runs closer to 3 characters per token, so a 60 KB source file is nearer 20,000 tokens than 15,000. Tables and JSON are denser still because punctuation and keys repeat. Japanese, Chinese and Korean can exceed one token per character on some tokenizers.

When the document is code or multilingual, treat the estimate as a lower bound and verify with the provider's tokenizer. The context-window calculator lets you switch the tokens-per-word assumption for code or mixed text.

FAQ

How many tokens is a page?

A 500-word page of English prose is about 665 tokens at 1.33 tokens per word, or roughly 2,000 characters. Code, tables and footnotes are denser; use a tokenizer for an exact count.

How many pages fit in a 200K context window?

About 300 pages of 500-word prose fit the raw 200K window. After reserving 8K for output, 2K for the system prompt and 2K for a margin, usable input is 188K, or about 282 pages.

Is the conversion exact?

No. The converter uses the common 1.33 tokens per word and 4 characters per token estimates. Prose varies by 10–15%, and code or non-Latin text can be further off. Use the provider's tokenizer when an exact count matters.

Why does my document fit the page count but fail the request?

The context window is shared with the output and the prompt scaffolding. A 199K-token document does not fit a 200K model once you reserve max_tokens, a system prompt and tool schemas. Subtract those before comparing.

Check the real context fit

The context-window calculator subtracts output, system prompt, tools and margin, then shows which models fit.

Open the context calculator → See models