How much do files actually shrink?

Median size reduction by file type

Sorted by median saving. Read the median column, not the mean — savings are heavily skewed, because a handful of badly exported files shrink enormously and drag the average up. The median is what a typical file of that type actually does.

What the numbers show

Already-compressed formats barely move

ZIP, WebP, plain text and GIF sit at or near a 0% median. Their bytes have already been squeezed by the format itself, so a second pass has almost nothing left to remove. If a tool promises to shrink a ZIP archive by 50%, it is either repacking it with a different algorithm or measuring something other than what you uploaded. Compressing these is usually not worth the upload time.

Uncompressed formats have the most to give

WAV audio and ICO icons store raw, unpacked data, which is why they lose the most — roughly nine tenths of the file. The same logic explains PNG screenshots: they are lossless by design, so re-encoding them with a smarter palette removes a lot.

PDF is the format most worth compressing

PDF is both the most-uploaded type here and a consistent performer, with a median around half the original size. That is because most oversized PDFs are oversized for one reason: images embedded at print resolution inside a document that will only ever be read on a screen.

The mean and the median disagree — on purpose

For several formats the mean sits well above the median. That gap is the story: a small number of very badly exported files shrink dramatically and pull the average up. Quoting the mean would make the typical result look better than it is, which is exactly how most published compression numbers end up so optimistic.

Method and limitations

• Every row is a real file uploaded by a user of micompress.com between 2026-05-06 and 2026-11-11. No synthetic test corpus, no hand-picked samples.
• Saving is measured as (original − compressed) ÷ original on the delivered file. Runs that failed, or where the result was not smaller, are excluded.
• Formats with fewer than 30 measured files are left out entirely rather than shown with a percentage that would say more about which files happened to be uploaded than about the format.
• These are self-selected files. People compress files they already suspect are too big, so this sample skews toward oversized files and the real-world figures are likely higher than you would get across a random assortment of files.
• Most results use the default "recommended" setting, which preserves visible quality. The "extreme" setting removes more; "lossless" removes less.
• Figures are recomputed automatically and were last updated 2026-11-11. They will drift as more files are processed.