NimboShort answer
Across 27 public PDFs (104.5 MB in total), Strong saved 49% of the total size, Recommended 27% and the lossless Light level 3%. Almost all of it came from photos stored at high resolution, in brochures and slide decks. Text documents and forms are already compact, and archive scans already use specialised compression, so the tool often hands the original back unchanged rather than a bigger file.
What we measured
We took 27 PDFs that anyone can download (880 pages): US tax forms, NIST reports and arXiv papers, NASA slide decks, USGS fact sheets, and catalogues printed between 1894 and 1920, scanned by libraries. Each one went through our Compress PDF page, the same code you use, at its three levels.
- Light saves the file again without touching its content: lossless, and the text stays selectable.
- Recommended and Strong redraw each page as a JPEG image (150 and 110 dpi). It is much smaller for photos, but the text can no longer be selected. When the images are not smaller than the Light version, the tool keeps the Light version.
- When a level cannot save at least 2%, the tool says the PDF is already optimized and gives nothing back. The tables show this as “unchanged”.
Results by type of document
| Type | Files | Light | Recommended | Strong |
|---|---|---|---|---|
| Forms | 2 | −2% | −2% | −2% |
| Text documents | 5 | −2% | −2% | −2% |
| Slide decks | 6 | 0% | −42% | −64% |
| Brochures and fact sheets | 6 | −17% | −61% | −80% |
| Scans | 8 | 0% | −4% | −32% |
| All files | 27 | −3% | −27% | −49% |
Brochures and fact sheets shrank the most: Strong saved 80% of their total size and Recommended 61%, because their photos are stored at print resolution. Slide decks come next (64% with Strong), but unevenly: most of it comes from one deck full of large photos, while several others did not shrink at all.
Scans were the surprise. Libraries such as the Internet Archive already compress their scans with formats designed for them (JBIG2 for text, JPEG 2000 for pictures), and every scan in our set used both. Redrawing those pages as ordinary JPEG images rarely helps: Recommended saved 4%, Strong 32%, at the cost of image quality. A scan straight from a phone or an office scanner is usually far less compressed.
Forms and text documents barely move (2% for text documents, whatever the level): they are mostly text and fonts, which are already compact. Redrawn as images they would get bigger, so the tool keeps the lossless version.
Over all 27 files, Light saved 3% of the total size, Recommended 27% and Strong 49%. Light made 5 files smaller, Recommended 11 and Strong 18: most PDFs published today are already well compressed, and the tool says so instead of returning a bigger file.
Speed and visual fidelity
At the Recommended level, the median processing time was 0.6 s per page, in headless Chromium 141 on a test machine (Intel Xeon, 2.80GHz, 4 cores). A recent laptop or phone should be in the same range; high-resolution scans take longer because every page is rendered and encoded again.
To see what the images cost, we compared the first three pages of each result with the original, both rendered at 96 dpi, using SSIM (structural similarity, where 1 means identical). On the 7 files that Recommended redrew as images, the median SSIM was 0.93; on the 15 files Strong redrew, it was 0.91 (lowest 0.81). In practice the text stays readable on screen and on paper, but it looks softer when you zoom in, and it can no longer be selected or searched.
Every file
| File (source) | Pages | Original | Light | Recommended | Strong | Time | SSIM |
|---|---|---|---|---|---|---|---|
| irs-form-1040Forms | 2 | 0.21 MB | unchanged | unchanged | unchanged | 1.3 s | — |
| irs-form-w9Forms | 6 | 0.13 MB | −5% 0.13 MB | −5% 0.13 MB | −5% 0.13 MB | 1.8 s | identical |
| arxiv-1512-03385Text documents | 12 | 0.78 MB | −13% 0.68 MB | −13% 0.68 MB | −13% 0.68 MB | 3.3 s | identical |
| arxiv-2101-00001Text documents | 11 | 0.16 MB | −18% 0.13 MB | −18% 0.13 MB | −18% 0.13 MB | 2.9 s | identical |
| irs-1040-instructionsText documents | 126 | 4.2 MB | unchanged | unchanged | unchanged | 24.2 s | — |
| nist-csf-2Text documents | 32 | 1.4 MB | unchanged | unchanged | unchanged | 4.4 s | — |
| nist-sp-800-63bText documents | 80 | 1.4 MB | unchanged | unchanged | unchanged | 9.3 s | — |
| nasa-crew-trainingSlide decks | 11 | 1.6 MB | unchanged | unchanged | −36% 1.0 MB | 2.1 s | — |
| nasa-dryden-overviewSlide decks | 29 | 20.3 MB | unchanged | −74% 5.2 MB | −88% 2.5 MB | 6.1 s | 0.974 |
| nasa-luvoir-conceptSlide decks | 29 | 1.9 MB | unchanged | unchanged | unchanged | 5.5 s | — |
| nasa-resource-prospectorSlide decks | 25 | 5.8 MB | unchanged | −3% 5.6 MB | −54% 2.6 MB | 6.3 s | 0.942 |
| nasa-sep-conceptsSlide decks | 14 | 1.4 MB | unchanged | unchanged | unchanged | 2.6 s | — |
| nasa-viper-roverSlide decks | 37 | 5.7 MB | unchanged | unchanged | −32% 3.9 MB | 5.8 s | — |
| usgs-amphibiansBrochures and fact sheets | 4 | 4.1 MB | −63% 1.5 MB | −63% 1.5 MB | −81% 0.77 MB | 3.6 s | identical |
| usgs-bee-labBrochures and fact sheets | 2 | 8.1 MB | unchanged | −90% 0.79 MB | −95% 0.36 MB | 4.5 s | 0.897 |
| usgs-fs-2016-3033Brochures and fact sheets | 4 | 1.5 MB | unchanged | −10% 1.4 MB | −59% 0.63 MB | 3.1 s | 0.914 |
| usgs-fs-2018-3022Brochures and fact sheets | 4 | 1.1 MB | unchanged | unchanged | −40% 0.66 MB | 3.8 s | — |
| usgs-satellite-photosBrochures and fact sheets | 2 | 0.34 MB | unchanged | unchanged | −14% 0.29 MB | 1.6 s | — |
| usgs-wetland-lossBrochures and fact sheets | 4 | 2.8 MB | −19% 2.3 MB | −36% 1.8 MB | −71% 0.83 MB | 2.2 s | 0.890 |
| scan-harkett-nursery-1894Scans | 54 | 9.3 MB | unchanged | unchanged | −37% 5.8 MB | 63.9 s | — |
| scan-hastings-seeds-1920Scans | 18 | 6.4 MB | unchanged | −25% 4.8 MB | −66% 2.2 MB | 28.7 s | 0.951 |
| scan-henderson-bulbs-1909Scans | 26 | 7.7 MB | unchanged | −3% 7.5 MB | −55% 3.5 MB | 44.9 s | 0.930 |
| scan-lineage-book-1900Scans | 168 | 8.4 MB | unchanged | unchanged | unchanged | 187.7 s | — |
| scan-pierson-seeds-1920Scans | 54 | 2.6 MB | unchanged | unchanged | unchanged | 48.5 s | — |
| scan-washington-calendar-1900Scans | 36 | 1.7 MB | unchanged | unchanged | unchanged | 35.2 s | — |
| scan-yachtsman-manual-1903Scans | 54 | 2.9 MB | unchanged | unchanged | −21% 2.3 MB | 56.5 s | — |
| scan-york-catalogue-1901Scans | 36 | 2.5 MB | unchanged | unchanged | −35% 1.6 MB | 31.2 s | — |
Download the measurements: JSON · CSV
What it means for your files
- A PDF that is mostly text (a contract, a form, a report) will not shrink much, and that is normal. Light is the safe choice.
- A document full of photos (a brochure, a catalogue, slides) is where compression pays off: try Recommended, then Strong if an upload limit forces you.
- A scan from a library or an archive may already be as small as it gets. A scan from your phone usually is not.
- To reach a precise size, use the “Target size” option, or follow our guide to get a PDF under 5 MB. Compression turns pages into images: to search the text afterwards, run OCR PDF on the compressed file.
How to reproduce it
The list of files with their sources, licences and SHA-256 fingerprints, every measurement and the script that drives the tool are published with this article. The files were measured on October 4, 2026. Nothing was uploaded anywhere during the test: the tool works inside the browser tab.