pdfpix

Best PDF compressor

Most compressors give you three quality presets and let you guess. A few let you name the size you actually need. If a portal demands a file under a stated limit, that difference decides whether the job takes ten seconds or ten attempts. This page ranks by that first, and by what each tool does to your text second.

Rank by whether you can name a number

A preset is a guess about your document. A byte target is a statement about the requirement. The upload form does not care whether you chose medium quality; it cares whether the file is under the limit, and only one of those two inputs answers that question.

The pdfpix compressor takes the target in kilobytes or megabytes and runs four methods against it: structural cleanup, image downsampling to 150 dots per inch with mozjpeg, font subsetting, and a raster fallback. The smallest candidate that passes verification is the one you get.

What the numbers actually look like

On a public test corpus of 120 cases the engine reached the requested target in 55 of them, 45.8 percent, with zero failures and no output larger than its source. Median size reduction was 59.2 percent and the ninetieth percentile 88.1 percent. Median time was 2512 milliseconds.

Reach depends heavily on the document. Scanned files reached the target in 88.9 percent of runs, mixed files in 55.8 percent, and text-only files in 20 percent. A text document is already small and there is little left to remove, which is a physical fact rather than a tool limitation.

The check nobody else publishes

Aggressive compression can flatten text into pictures. The file gets small and stops being searchable, which nobody notices until someone tries to find a name in it. Every pdfpix candidate is read back and its text compared to the source before it can be offered.

The floor is ninety eight percent of non-space characters, and on the public corpus text retention was 1.0000 at the median, the tenth percentile and the minimum. A candidate that fails is discarded, not shipped with a warning.

The free tools people try first

iLovePDF offers quality presets and processes on its servers; its free tier is described as essential tools with limited document processing, with no number published. The comparison covers it.

Smallpdf marks strong compression and batch compress as unavailable on its free plan, checked 20 August 2026. PDF24 states it is free without restrictions, funded by advertising, and processes its web tools on its servers.

How to run a fair test yourself

Take one real file, not a sample, and put the same file through each candidate at the same target. Then check three things: the resulting size, whether the text is still selectable, and whether the images are still readable at the zoom your reader will use.

Rankings that skip that step are ranking marketing copy. Ten minutes with one document tells you more than any list, including this one, and the file you test should be the kind you actually send.

Questions people ask

Which compressor lets me set an exact file size?
The pdfpix compressor takes a target in kilobytes or megabytes and works toward it with four methods. Most free tools offer three quality presets instead, which means retrying until the result happens to fit.
How small can a PDF get without losing the text?
It depends on what is in it. On the public test corpus the median reduction was 59.2 percent and the ninetieth percentile 88.1 percent, with selectable text intact in every run. Scans shrink most; text-only documents shrink least.
Why do compressors give different results on the same file?
They optimise different things: image resolution, image quality, font subsetting, object cleanup, or flattening the whole page to a picture. The last one produces the smallest file and destroys the text, which is why the output needs checking.
Does compressing a PDF make the text unsearchable?
It can, if the tool flattens pages into images. pdfpix compares the output text against the source and requires ninety eight percent of non-space characters to survive, discarding any candidate that fails rather than offering it.

The tools this compares