The Best Image Compressor for Mac in 2026

Choose Smol when shrinking images is part of your work: product photos for a shop, assets for a website, or a folder of client exports. You get batch compression, conversion and metadata removal in one window, and the same app handles PDFs, video and audio. There is no monthly quota and nothing is uploaded.

1. Drop in your images or a folder.
2. Choose JPEG for broad compatibility, or WebP or AVIF for the web, and set quality.
3. Compress the batch and check the results before publishing.

What you are buying is output on a level with the best free tool we measured, for a whole folder at once. On JPEG and on WebP, Smol and Squoosh needed the same bytes for the same score across the 24 test images. Squoosh does one image at a time. Smol did 24 in 7.3 seconds.

Against ImageOptim, which most lists put first, the difference is the format. With ImageOptim’s lossy mode switched on, its JPEGs and Smol’s JPEGs came out level at the same score. Smol’s AVIF needed 20 to 22% fewer bytes than ImageOptim’s JPEG for that score, and 71 to 76% fewer than ImageOptim’s PNG when the photos started as PNG. ImageOptim cannot write AVIF or WebP, or convert anything. The numbers are in its entry below.

Against Apple’s built-in encoder the picture is closer than we would like to claim. At web quality Smol’s JPEGs needed 7 to 10% fewer bytes than sips for the same score. At high quality, a score of 85 or about our quality 90, sips was 6% ahead. The larger win is AVIF: Smol’s needed 15 to 21% fewer bytes than Apple’s AVIF at the same SSIMULACRA2 score, and 22 to 26% fewer than its own JPEG.

Defaults matter in a batch tool, so here are ours. Metadata is stripped: we checked a Sony camera file and every EXIF field was gone from Smol’s output while all of them survived sips. Chroma is 4:2:0 up to quality 75, 4:2:2 at 80 and 85, and full 4:4:4 from 90, which is why quality 90 costs noticeably more bytes than 85. Images are also resized to fit inside 2000 × 2000 px unless you change that, so raise the limit for large originals.

There is also a file-size target. We asked for an AVIF of 20 KB or less from the 557,596-byte kodim23 and got 10,182 bytes, inside the requested band. Two caveats. The target is a band, not a ceiling, and it is slow: around 28 seconds for one 768 × 512 image, against a third of a second for ordinary compression. It works toward a byte budget rather than a quality score, so every number in this article comes from ordinary compression.

Smol is $29 once for one Mac, or $49 for a team pack of three, with updates included and no subscription. There is no free trial and purchases are not refundable. Photographers and shops can read how it fits their work in Smol for photographers and Smol for e-commerce. The document side is measured separately in our PDF compressor roundup.

ImageOptim is a window over the best open-source optimizers: it ships oxipng, zopflipng, pngout, advpng, pngquant, gifsicle, jpegoptim, guetzli, svgcleaner and a MozJPEG build of jpegtran. Drop files in, get the same pixels in fewer bytes. We verified the lossless claim instead of repeating it, and it holds: an absolute pixel error of zero on every one of our six outputs, including the JPEG.

Where it beats everything else on this page is flat graphics. Our transparent badge went from 101,316 to 45,626 bytes, a 55% cut, with nothing lost. If you ship UI assets, diagrams or logos and the pixels must not change, this is the tool and it costs nothing.

Photographs are another matter. On 9 October we ran version 1.9.3 over the full 24-image Kodak suite. At its defaults it took 4.3% off the PNGs and 8.9% off the same photos saved as JPEG. That is about what lossless work can do for a photo (Caesium’s lossless mode managed 4.5% on the same PNGs), and it is why people drop a folder of pictures on ImageOptim and conclude it did nothing.

It does have a lossy mode, off by default, in Preferences. With it on at the default quality of 80 the JPEGs came out 63.8% smaller in under a second, and matched score for score they were level with Smol’s JPEGs. So for JPEG in, JPEG out, ImageOptim is as good as anything here and faster than we are. What it cannot do is leave JPEG and PNG behind. From the same files, Smol’s AVIF needed 20 to 22% fewer bytes than ImageOptim’s JPEG at the same score, and 71 to 76% fewer than its lossy PNG. Lossy mode also rewrites your files in place and moves the originals to the Trash, so fish them out before you empty it.

Its release situation is odd. The shipped build is 1.9.3, dated 29 October 2023 by its own Sparkle feed, yet the repository is alive: the latest commit to main landed 13 July 2026. That commit is PR #466, which stops ImageOptim pinning its worker processes to Apple Silicon efficiency cores, and it has never shipped. We confirmed the behavior: the bundled defaults.plist sets RunLowPriority to true, and ps -axo pid,nice,args showed every optimizer at nice 10.

The advertised workaround, defaults write net.pornel.ImageOptim RunLowPriority -bool false, changes nothing measurable on 1.9.3: the helpers stayed at nice 10, and three timed trials came out at 5.23 s median with it on against 5.52 s with it off. Speed is the practical cost: our six-file batch took 99.26 s. Its bundled oxipng is 9.0.0 against an upstream v10.2.1 released 2 September 2026.

The full head-to-head, with every setting and timing, is on our ImageOptim alternative page and in the benchmark data.

For a single image, File → Export in Preview with the quality slider is a perfectly good answer and we are not going to pretend otherwise. For anything repeatable, sips is the underrated one: sips -s format jpeg -s formatOptions 60 in.png --out out.jpg works in any script, on any Mac, with nothing installed. A loop over our 24 test images finished in about half a second.

Its JPEG encoder is respectable. To match Smol’s score across the suite it needed 8 to 12% more bytes at web quality, and at high quality it was 6% smaller. Nobody should buy software over a gap that size.

On macOS 27 it also writes AVIF, which almost nobody knows. sips -s format avif -s formatOptions 75 produced a valid 46,967-byte AVIF from kodim07. Run sips --formats and read the Writable column yourself. Apple’s AVIF tied ours on SSIM and trailed on SSIMULACRA2, where it needed 18 to 26% more bytes for the same score.

Three sharp edges. WebP is listed as readable but not writable, and asking for it fails silently: sips -s format webp prints the file paths, exits 0, and writes no file. The quality scale is its own thing: formatOptions 50 lands where Smol’s quality 80 does, and the named preset best switches to 4:4:4 chroma and made our test file three times larger than formatOptions 95. And it keeps your EXIF. Camera model, timestamp and GPS all survived in our output, which matters if you are about to upload holiday photos.

If the loop is the part you would rather not write, a Finder Quick Action gets you right-click compression without Terminal.

Nothing else here lets you drag a divider across the image and watch the artifacts appear in real time, with two codecs either side and a live byte count. When you are deciding whether a hero image can survive AVIF at a low setting, Squoosh answers in ten seconds and every desktop app on this list answers in export-and-check cycles. It runs entirely in the browser, so nothing is uploaded.

We drove the web app in headless Chrome to get real numbers out of it. Across all 24 images, its JPEG and WebP output and Smol’s needed the same bytes for the same score. If you like what Squoosh does to one image, Smol does that to a folder. Its lossless OxiPNG pass saved 4.1% on the suite.

The maintenance picture needs stating precisely. The last commit to the default branch was 19 August 2024. In issue #1472, opened 18 May 2026 and titled “Abandoned?”, Jake Archibald replied the same day: “The problem is the folks who built it left Google, and Google refused to hand the project over, or maintain it. That said, I still have commit & publish access, so I was planning on updating the JPEG XL and AVIF codecs some point soon.” As of today the site is up and works.

The command-line build is the reason Squoosh cannot be your batch tool. @squoosh/cli was last published on 3 January 2023 and declares support for Node 12, 14 and 16 only. We installed it on Node v26.7.0 and it threw before encoding anything: TypeError: Cannot set property navigator of #<Object> which has only a getter.

Alternatives are laid out in our Squoosh alternatives comparison.

TinyPNG earned its reputation on PNG quantization, and the free web tool remains the fastest way to make an image smaller on a computer that is not yours. Drop files on the page, download the results. Its current limits, from its own page, are up to 20 images at a time, 5 MB each, with format conversion limited to 3 images. It handles JPEG XL, AVIF, WebP, JPG, PNG and APNG.

The part worth paying for is the API. Tinify gives 500 compressions a month free with no payment method, then charges $0.009 per compression for the next 9,500 and $0.002 for the 10,000 after that. If you want compression inside a deploy step and not on a laptop, that is a reasonable way to buy it.

The trade-off is the obvious one and it is not small: your images go to somebody else’s servers. For product photography that is fine. For client work under NDA, medical images, unreleased designs or anything with faces and GPS coordinates in the EXIF, it is a policy decision before it is a technical one. The 5 MB and 20-file ceilings also rule out a folder of camera exports. We did not measure it: there is no way to run it on a set of files without sending them to TinyPNG.

The head-to-head is in Smol vs TinyPNG.

Zipic sits at almost exactly our price and covers almost exactly our ground: drag and drop, batch, presets, local processing. Its site lists Personal at $29.99 for two devices and a Team license at $129.99 for ten, both one-time with lifetime updates, and a free tier that runs a seven-day Pro trial then settles at 25 images a day. Version 1.10.2 is a universal binary for macOS 13 or later, and there is a Windows beta.

It beats us on format breadth in a direction that matters to some people. Its Pro formats include JPEG XL, SVG, APNG and ICNS, and Smol writes none of those. It also ships a proper command-line tool inside the app bundle, which is how we tested it.

We ran the free tier on three of the Kodak images, which is what its daily limit allowed alongside a sweep of its six levels. At level 3, its balanced preset, kodim07 came out at 47,104 bytes; Smol at quality 75 produced 44,093 bytes at the same SSIM and SSIMULACRA2 scores, 6% smaller. Tuned to equal scores of 70 and 75 across the three images, Zipic needed 4 to 6% more bytes than Smol and beat sips. At its level 2 the two were level, and at its highest-quality level it was ahead of us, by about 10% on the one image we tested there. Call it a draw on JPEG.

The free tier is narrower than the feature list suggests. Asking the CLI for AVIF returned “Converting to AVIF is a Pro-only feature”, and --no-metadata returned “Removing image metadata is a Pro-only feature”. If you plan to stay on the free tier, your EXIF stays in your files.

Caesium is the free answer for people who want a lossy compressor with a window and a slider. The desktop app is version 2.8.5, released 17 May 2025, for macOS 12 or later; its engine, libcaesium, is actively developed, and the command-line build caesiumclt 1.5.0 came out on 16 September 2026. We measured the command-line build from Homebrew.

Two results stood out. On WebP it was level with Smol and Squoosh, so there is nothing to choose between the three. And it is the fastest tool here: the whole suite to JPEG in 0.2 seconds.

JPEG is where it falls behind. It needed 13% more bytes than Smol at a score of 70, the gap widens from there, and it has a ceiling: at quality 95 its JPEGs averaged a SSIMULACRA2 score of 76 and never reached 80 on any of the 24 images, where Smol and sips go well past it. For web thumbnails that is plenty. For photographs you care about, it is the wrong encoder. Lossless mode saved 4.5% on the suite.

The longer comparison is in our Caesium alternatives page.

Compresto covers the same span we do: images, video, GIFs and PDFs, with a drop zone and folder monitoring, processed locally. Its pricing page lists Personal at $49 for one Mac and Standard at $69 for two, one-time with lifetime updates, and it is also sold as a subscription.

The useful reason to choose it is Setapp. If you already pay that subscription, Compresto is included, and paying $29 or $49 again for a second app that does the same job is hard to justify. We would not.

Its documentation lists JPG, PNG, WebP and AVIF as output formats, read from JPG, PNG, WebP, AVIF, HEIC, TIFF, BMP and static GIF, with five named quality levels in place of a number. We did not run it, so there are no Compresto numbers on this page and there will not be any until we can test it properly.

JPEGmini has one idea and has had it for years: recompress a JPEG as far as it will go before a viewer could tell, with no slider to get wrong. Pro is $59 once for three devices with a year of updates; plug-ins for Lightroom, Photoshop and Capture One take it to $89, and watch folders to $99. The free tier is a seven-day trial, then 50 images a day.

The limit is in the name. It reads JPEG and HEIC and writes JPEG. On our test images a change of format was worth 22 to 26%: that is how many fewer bytes Smol’s AVIF needed than the best JPEG we measured at the same score. If your deliverable is always a JPEG for a client or a print lab, JPEGmini’s no-decisions workflow is a real advantage. If your images are going on a website, it cannot write the formats that would help most. We did not measure it.

Scope and price are compared in our JPEGmini alternatives page.

Optimage’s idea is good: in place of a quality slider it targets a perceptual quality level and finds the smallest file that meets it, which is roughly the right way to think about compression. At $15 one-time, with a free tier capped at 24 images a day, it is also the cheapest paid app here.

The problem is architectural and it has a date on it. We downloaded the current build from optimage.app, unpacked it, and ran lipo -archs across every Mach-O binary inside. All thirteen, including the main app and its bundled ffmpeg, Ghostscript, node and cwebp, report x86_64 and nothing else. There is no arm64 slice, so on every Apple Silicon Mac it runs under Rosetta.

Apple announced on 1 September 2026 that “macOS 27: Final release to support Rosetta”, and that Intel-only apps will no longer run on Apple Silicon Macs after that update. The app’s own version history page has no entry newer than 26 November 2021. Buy it if you want, but you are buying software with a published expiry date. The migration options are in our Optimage alternatives comparison.