The short version

Recommended, Strong and Extreme recompress the images inside the file in place and never touch the text, so it stays selectable and searchable. Lossless only tidies the structure. Scan rebuilds each page as a picture with an invisible text layer, which is what a scanned document needs.

A before and after preview of page one appears under the result, so you can judge the trade before downloading. The output is never larger than the input: if a level cannot help, the original comes back and the result says so.

Every PDF that will not fit through an inbox or a portal is oversized for a reason that can be named. Usually it is images: a phone scan stores each page as a photograph, and a photograph of an A4 sheet at 300 dpi is several megabytes before anything else is counted. Sometimes it is fonts, or a bloated structure left behind by repeated saves. Compression only works when the mode you pick matches the reason, which is why this page starts with a diagnosis rather than a button.

How to compress a PDF in your browser

1

Add the PDF

Drop it or browse. The file is read into memory; nothing is uploaded, so a 200 MB scan on a slow connection is no different from a small one.

2

Pick a level

Recommended is right for almost everything. Strong and Extreme go further when a cap demands it. Lossless changes nothing visible. Scan is for documents that are already pictures of pages.

3

Tick greyscale or metadata removal if they apply

Greyscale halves the weight of a colour scan of a black and white page. Metadata removal clears the author, title and XMP record along the way.

4

Press Compress PDF and read the result

The result line shows the exact before and after, and page one is drawn at both sizes underneath. If you overshoot a cap, pick the next level and run it again on the original, not on the output.

Try it

Which mode should you use?

Three questions, one recommendation. Based on what is in the file, whether you need the text afterwards, and how small it has to get.

What each level does

Every level except Scan works the same way: the tool walks through the images embedded in the file, decodes each one, brings it down to a resolution the page can actually use, encodes it again as a JPEG and writes it back in place. The page's text, vector drawings, links, bookmarks and form fields are never rewritten. That is how Acrobat, iLovePDF and PDF24 compress too, and it is why a compressed contract still lets you search for a clause.

The levels differ only in how far the images are pushed. Recommended keeps 150 pixels per inch, which is more than a screen can show and enough for an office printer. Strong drops to 110. Extreme goes to 72, the point where photographs start to look soft even on screen. Lossless skips the images entirely and only tidies the structure: thumbnails, private application data and, if you tick it, metadata.

Recommended on a photo report
40 to 75%

Images recompressed at 150 ppi. Text, links and forms untouched.

Scan on a phone scan
60 to 90%

Each page rebuilt as a JPEG at 150 dpi with a searchable text layer.

Bytes uploaded
0

Every level runs in the browser with pdf-lib and PDF.js. The file never leaves the device.

Not every image can be recompressed safely. Pictures stored as JPEG 2000, fax-style bilevel scans, CMYK print images and anything with an unusual colour setup are left exactly as they were rather than risked. The result line tells you how many images were changed and how many were left alone.

Scan mode, and what it costs

Scan takes each page, renders it with PDF.js at the resolution you pick, encodes the result as a JPEG and replaces the page with that picture. That is the same thing a scanner does, which is why it works so well on scans: a page that was already a picture loses nothing structurally and gains a far more efficient encoding. Unlike a scanner, the tool then writes the page's text back as an invisible layer, so search, select and copy keep working.

Inside Scan modeFour steps, no server
Pagetext, vectors, imagesRender100, 150 or 200 dpiJPEG encodeone image per pageText layerinvisible, searchableEvery page becomes a picture with the words written back underneath
The output prints and displays like any PDF, and a search still finds the words. Vector drawings and small text are the only things that visibly change.
Scan is the wrong choice for a text document
A page that was crisp vector text becomes a photograph of itself. The text layer keeps it searchable, but zooming in shows pixels, and vector diagrams or CAD drawings lose their edges. Use Recommended or Strong on anything that came out of Word, a browser or a design tool, and keep Scan for pages that were already pictures.
Resolution by resolutionWhat each Scan setting looks like
  • 200 dpiPrints cleanly, 40 to 60% smaller than a 300 dpi scan
  • 150 dpiClean on screen and fine for a desk printer, 60 to 80% smaller
  • 100 dpiReadable, visibly soft, 75 to 90% smaller

Which files shrink, and which do not

Strong works hard on

Files whose weight is pixels

  • Phone or copier scans, colour or greyscale
  • Reports full of screenshots, photos and diagrams
  • PDFs exported from image editors at 300 dpi or more
  • Slide decks with full-bleed photography
  • Forms that were printed, filled by hand and scanned
Neither mode moves much on

Files that are already lean

  • Text-only PDFs from Word, Docs or LaTeX
  • Output from another compressor, already at low quality
  • Files whose size is embedded fonts you need
  • Vector drawings and CAD exports (Scan makes them worse; the other levels leave them alone)
  • Files that are small already: 300 KB rarely needs help

One rule of thumb resolves most doubt. Open the PDF and try to select a sentence. If the cursor grabs words, it is a text document: Recommended will shrink the images inside it and leave the words alone. If it selects a whole page as a block, it is a scan, and Scan mode is the level that will make a real difference.

Hitting a specific size

Most limits are round numbers set years ago and never revisited. The settings that usually clear them, for a typical mixed document:

Common ceilingsWhere to start
  • 25 MB, Gmail attachmentLight, then Strong Balanced
  • 10 MB, Outlook and ExchangeStrong Sharp for scans
  • 5 MB, court and grant portalsStrong Balanced
  • 1 MB, job and visa portalsStrong Balanced or Small
  • 500 KB, school and NHS formsStrong Small, few pages
  • 300 KB down to 50 KB, ID and portal uploadsA dedicated target page

For the tighter ceilings the guesswork is removed entirely. The target pages, 50 KB, 100 KB, 150 KB, 200 KB, 300 KB, 500 KB, 1 MB, 2 MB, 5 MB and 10 MB, work through three stages, tidying the file, then recompressing the pictures inside it, then rebuilding the pages only if they must, and keep the first result that fits. For inboxes there are email, Gmail and WhatsApp variants that account for the encoding overhead mail servers add.

Compress once, at the end
Merge first, stamp and sign first, compress last. Compressing the pieces and then merging gives a worse result than merging and compressing once, and compressing an already-compressed file compounds the quality loss without much further saving. Keep the original and produce a compressed copy for each destination.

The caps the portals actually enforce

Most people arrive at a size target because an upload box has just refused a file. These are the limits that come up most, checked in September 2026. Every one of them is set by the portal rather than by the PDF format, and every one of them has moved at some point, so treat the number as a starting point and confirm it against the form in front of you.

Portal and capLimits worth knowing before you compress
  • UKVCAS, UK visa applications made in the UK6 MB per file
  • VFS Global, UK visa applications overseas2 MB per file
  • TLScontact, UK visas in some countriesabout 4 MB
  • USCIS online filingvaries by form, 6 MB upward
  • Federal court e-filing through CM/ECFset per district, about 15 to 75 MB
  • Workday and similar job portals5 MB stated, often less enforced

Three of these behave badly enough to be worth calling out. UKVCAS is the worst: a file over the limit does not produce an error, it simply fails to attach, so applicants finish the session believing the evidence went up. GOV.UK guidance on uploading evidence covers what the Home Office expects, and the practical answer is to compress and check the file list before you leave the page.

Job portals built on Workday state one figure and enforce another. The upload box may accept 5 MB while the parser behind it is configured lower, which means the file uploads, the confirmation appears, and the resume never reaches the searchable database. Under 2 MB avoids the question entirely, and taking a photograph out of a CV usually gets you there on its own.

Federal court filing has no single number at all. Each district sets its own limit in CM/ECF and they now range from about 15 MB to 75 MB, having risen steadily as storage got cheaper. PACER's answer on file size explains how to find the figure for a given court rather than guessing.

Immigration filing works the same way. The ceiling depends on which form you are submitting rather than on the account, so USCIS filing tips is the place to confirm it before you start compressing towards a number somebody quoted on a forum.

Where a document will not fit under any of these, splitting it is almost always accepted and almost always better than compressing until the page stops being readable. Most of these systems take several attachments, and two legible files beat one unreadable one at every counter that will ever look at them.

Why the file never leaving your device matters here

The PDFs people most need to shrink are the ones they least want to hand around: bank statements for a lender, medical records for a referral, a signed contract that will not fit through the counterparty's mail gateway. Most online compressors upload the file, process it on a server, and email back a link. For those documents that is a real exposure for the sake of a smaller attachment.

Two ways to compressThe browser path never transmits the document
Your devicePDFin memoryBrowser enginecompressDownload0 bytes leavethe machineTypical online toolPDFon diskuploadServercopies filedownloadA full copy now sits on hardware you do notcontrol, under a policy you cannot audit
PDF.js and pdf-lib run as JavaScript inside the tab. The file is read into memory, rewritten or re-rendered there, and offered as a download. Close the tab and it is gone.

Checking the result

  • Open it and zoom to 200 percent on a page with the smallest text. If it reads cleanly, the setting was right. If not, run the original again at the next quality up.
  • Confirm the page count in the result line matches the source. It always should, and the check takes a second.
  • Try to select text. Every level keeps it, including Scan, which writes the words back as an invisible layer.
  • Still too big? Split the file and send it in parts, or use the target page for the exact ceiling you are facing.

The tool is right above

Everything on this page runs in your browser. Scroll up, drop the file, and keep it on your machine.

Jump to the tool

Common questions

What people ask once they see the before-and-after number.

01How much can compression shrink a PDF?

Photo-heavy files typically lose 40 to 75 percent at Recommended and scans 60 to 90 percent in Scan mode. Text-only PDFs with no images are already efficient and may shrink only a few percent, or not at all.

02Does compression reduce quality?

Lossless does not. Recommended, Strong and Extreme recompress images, which softens photographs slightly at Recommended and visibly at Extreme, and never touch text. Scan re-renders the page, so text becomes pixels on screen even though it stays searchable.

03Which mode should I use?

Recommended first, for anything. If the file is still too big, Strong, then Extreme. If the pages are scans, Scan at 150 dpi. The picker above turns three answers into a recommendation.

04Why did Light say there were no savings?

Because the file has no images to recompress and was already well packed. Modern PDF writers use the same structural tricks Lossless applies, so there is nothing left to remove. That is honest reporting; the tool hands the original back rather than a larger copy.

05Can I compress a password-protected PDF?

Not while it is encrypted, because the pages cannot be read. Remove the password from a file you own with the Unlock tool, compress, then protect the result again if needed.

06What is the biggest file the tool can handle?

There is no policy limit. Browser memory is the ceiling, and Scan mode needs room to render pages, so files above a couple of hundred megabytes are better split first or run on a laptop rather than a phone.

07Will the compressed file open everywhere?

Yes. Both modes produce a standard PDF 1.7 that opens in Acrobat, Preview, every browser and every mobile reader, and uploads to signing services and portals without complaint.

08Is it free?

Free. Every tool, unlimited use, no signup.