How Many Tokens Is a PDF? (And How to Shrink It)

If you send documents to ChatGPT or Claude, tokens are what you pay for and what fills the context window. So a fair question is: how many tokens is a PDF? The honest answer is that it depends, and the gap between a clean and a messy version is large.

What decides a PDF's token count

• Amount of text: a dense 30-page report is far more than a 2-page memo.
• Tables and numbers: numeric, symbol-heavy content tokenizes less efficiently than prose.
• Extraction quality: broken words, artifacts and repeated page furniture inflate the count.
• Language: some languages use more tokens per word than English.

As a rough feel, a page of dense text is often somewhere in the hundreds to low thousands of tokens, so a long report can easily run past 20,000. But the same document, cleaned, can be much smaller.

Why messy extraction costs extra

When text is extracted badly, you pay for waste. A word split by a line break becomes two tokens instead of one. Repeated headers and footers add up across every page. Encoding artifacts like (cid:32) are pure noise you are billed for. This waste is commonly 30 to 60 percent of a raw document's tokens.

How to count accurately

Do not estimate from word count. Use a real tokenizer, the same one your model uses, on the actual text you plan to send. Word-to-token ratios vary too much to guess, especially with tables and numbers.

How to shrink it

1. Strip repeated headers, footers and page numbers.
2. Rejoin words broken across lines.
3. Convert tables to compact Markdown instead of padded layouts.
4. Send a cleaned, compact version and cache it so you do not re-parse every call.

PackForAI does this and shows the token count before and after on every file, so you can see the reduction directly (often up to 65 percent). Upload a PDF at packforai.com and read your document's real token count.

How many tokens is a PDF? Fewer than you are probably sending, once you strip the waste the model never needed.