How Much Smaller Can a PDF Get? Results From 98 Real Files
Keeping the text, a typical PDF got a quarter smaller; turning pages into images, nearly two thirds. Text-only files did not shrink at all. Here is what decides it.
Every PDF compressor promises a smaller file, but how much smaller depends almost entirely on what is inside the PDF. To put numbers on it, in November 2026 we ran 98 real, public PDFs through our PDF Compressor, with six different settings each, and measured every result. The files had a median size of 6.6 MB and 23 pages.
The files we tested
| Type | Files | Where they come from |
|---|---|---|
| Scanned books and newspapers | 35 | Internet Archive |
| Conference slide decks | 16 | Wikimedia Commons (Wikimania talks) |
| Reports and fact sheets with photos and maps | 22 | U.S. Geological Survey |
| Research papers with many figures | 15 | arXiv (computer vision) |
| Research papers that are almost only text | 10 | arXiv (mathematics) |
Results: median saving per file type
Keep text recompresses only the pictures inside the PDF, so text stays sharp and selectable. Maximum redraws every page as one image, which is smaller but turns the text into a picture.
| Type | Keep text, Recommended | Keep text, Strong | Maximum |
|---|---|---|---|
| Scanned books and newspapers | 24% | 41% | 70% |
| Slide decks | 30% | 48% | 56% |
| Reports with photos | 24% | 25% | 69% |
| Papers with many figures | 62% | 73% | 55% |
| Text-only papers | 0% | 0% | 0% |
| All 98 files | 25% | 40% | 63% |
Three things stand out. First, text-only PDFs do not shrink: in 8 of the 10 mathematics papers no setting saved anything, because text and fonts are already stored compactly. If such a file is too big, it is usually because of embedded fonts or a few large images, not the text. Second, papers full of figures shrank more with Keep text than with Maximum, because their images were stored at far more detail than a screen needs, while their pages are mostly white space and text that Maximum has to draw as pixels. Third, Strong barely helped the reports (25% against 24%): their pictures were already compressed JPEGs, so lowering the quality again gains little.
Under 1 MB: realistic. Under 100 KB: rarely
We gave every file of up to 80 pages a size target. The tool tries the gentlest settings first and turns pages into images only when it must.
- 1 MB: of the 72 files that were above 1 MB, 70 got under it. Only 15 of them managed it while keeping the text; the rest needed Maximum. The two misses had 53 and 56 pages.
- 100 KB: 5 of 84 files got under it, and all five had 5 pages or fewer. A longer document simply has too many pages: even drawn at low resolution, each page costs several kilobytes.
So if an upload form asks for 100 KB, the practical answer for a long document is to send fewer pages: delete the pages the form does not need, or split the file, before compressing.
Why scans behave differently
34 of the 35 scans from the Internet Archive store their pages as JPEG 2000 pictures, often combined with JBIG2, a black-and-white format for text. Browsers cannot read JPEG 2000 by themselves, and before this test our Keep text mode left those pictures untouched, so those scans saved nothing in that mode. The test made us fix it: Keep text now decodes JPEG 2000 with the same PDF engine that draws the pages and recompresses it like any other picture, and the numbers above were measured after the fix. JBIG2 is left as it is, since it is already about as small as a page of text gets.
Scans are also the slowest files. With Maximum, a scanned page took a median 2.7 seconds to redraw, against 0.4 seconds for a typical page overall, and Keep text took a median 47 seconds per scan. A 300-page scanned book takes a while in any browser tool.
What this means for your file
| Your PDF | What to expect |
|---|---|
| Mostly text, exported from Word or LaTeX | Little or nothing to gain; check for large embedded images instead |
| Slides or reports with photos | About a quarter to a half smaller with text kept |
| Paper or brochure with large, detailed images | Often more than half smaller with text kept |
| Scanned document | A quarter to 40% with text kept; about 70% as images |
| Must fit 1 MB | Usually possible up to a few dozen pages, often only as images |
| Must fit 100 KB | Only realistic for a few pages |
How we tested
- Each file went through the PDF Compressor page itself, in Chrome 157 without a visible window, driven by a script that clicked the same controls a person would. Nothing was uploaded anywhere; the tool runs in the browser.
- Settings: Keep text at Low, Recommended and Strong; Maximum at Recommended; and a target of 1 MB and of 100 KB. A result counts as saving nothing when the tool kept the original because its output was not smaller.
- Size targets were run on the 84 files of up to 80 pages; on the 14 longer ones each attempt redraws hundreds of pages, and a 100-page file is not what people squeeze into an upload form.
- We describe a file as text-only or photo-heavy from its own contents: how much of it is pictures, and in which format they are stored. Those figures are in the data file too.
Every file, setting and result is in one file, with a link to each source PDF: download the data (CSV).