Image compressor benchmark: Smol’s AVIF was 22 to 26% smaller than JPEG at the same quality

AVIF, by a wide margin. Smol’s AVIF was 21 to 24% smaller than its WebP and 22 to 26% smaller than its JPEG at the same score. WebP against JPEG was close: from 3% larger to 6% smaller, depending on the score.

These are one encoder’s three formats on film scans. Browser support, editing apps and upload limits decide which format you can ship; the full encoder-by-encoder table is below.

A compressor can always make a smaller file by keeping less of the picture, so file size alone says nothing. Each encoder was run at several settings on every image, each output was scored against its original, and we then worked out how many bytes each encoder needs to reach the same score. Totals are bytes for all 24 images; the percentage is the gap to Smol’s AVIF.

SSIMULACRA2 is a perceptual quality metric from Cloudinary that ships with the JPEG XL reference software; its authors’ README describes a score of 70 as high quality, 80 as very high quality and 90 as visually lossless. “n/a (2 of 24)” means none of the settings we ran reached that score on 2 of the 24 images, so no fair total exists. Caesium’s WebP is level with Smol’s; its totals differ slightly because we ran it at more settings, which gives the interpolation more points and a wider range.

The older SSIM metric tells a flatter story. By SSIM the two AVIF encoders are 2% apart, and Apple’s JPEG encoder beats Smol’s:

Squoosh is the free reference most people compare against, and for two of three formats there is nothing to separate them: on JPEG and on WebP the two need the same bytes for the same score. Smol is not ahead of Squoosh on either. What it adds is a whole folder at once.

AVIF differs. We ran Squoosh’s AVIF at its default setting only, then found the bytes Smol’s AVIF needs to match each file’s score:

The two metrics disagree, and we report both. An image is “comparable” when Squoosh’s score falls inside the range of settings we ran for Smol.

Added 9 October 2026. ImageOptim is the free Mac optimizer most lists still recommend first, so it gets its own test. It keeps whatever format it is given and writes neither AVIF nor WebP, so we ran version 1.9.3 from two starting points: the 24 original PNGs (15,394,305 bytes) and the same images saved as JPEG with sips at quality 90 (4,082,179 bytes). Its default is lossless; its lossy mode is a switch in Preferences with quality 80 as the default. Scores are against the original PNG in every case.

“Unchanged” means lossless: a pixel comparison found no differences in any of the 48 files. Left at its defaults ImageOptim took 8.9% off the JPEGs and 4.3% off the PNGs. Times are the mean of three runs, listed in the README.

Fixed settings do not say which tool is more efficient, so here is the same-quality comparison, with ImageOptim’s lossy mode on and both tools run at several settings:

Kept in the same format, the two are level: JPEG to JPEG runs from Smol 1% smaller to ImageOptim 4% smaller, and lossy PNG to lossy PNG is within 0.5%. The gap is the format. Smol’s AVIF needed 20 to 22% fewer bytes than ImageOptim’s JPEG and 71 to 76% fewer than its PNG, and ImageOptim cannot write AVIF. By SSIM the AVIF gap from JPEG is 34 to 35%, and JPEG to JPEG tilts to ImageOptim, whose files are 1 to 8% smaller. Smol’s WebP was 4 to 6% larger than ImageOptim’s JPEG at the same SSIMULACRA2 score.

On speed the result splits. ImageOptim finished the JPEGs about 8 times faster than Smol. On PNGs, Smol at the setting that matches ImageOptim’s lossy output in size and score (PNG at 90, 5,444,974 bytes at 81.7) took about 10.4 seconds, which is 1.6 times faster than ImageOptim’s lossy mode. ImageOptim’s own pull request #466, merged in July 2026 and not yet released, describes the shipping app pinning its work to Apple Silicon efficiency cores; we tested the released build and not the fix. What this means when choosing between the two is on the ImageOptim alternative page.

Wall-clock time to convert all 24 PNGs to one format, on an M2 Pro.

Smol is the slowest of the three by a wide margin, and we did not measure where the time goes. Squoosh is not in this table because it has no batch mode. ImageOptim’s times are in its own section, because it was run on a different day and on two kinds of input.

The 24 source PNGs are 15,394,305 bytes. Smol has no lossless mode, so it is absent here. ImageOptim’s row is from the 9 October test.

• Speed. A sips loop was about 13 times faster and caesiumclt about 37 times faster.
• JPEG at high quality. At a SSIMULACRA2 score of 85, Apple’s JPEG encoder needed 5.8% fewer bytes than Smol’s. By SSIM, Apple’s JPEG was smaller at both scores we matched.
• AVIF, by SSIM. Squoosh’s default AVIF was 4.2% smaller than Smol’s at the same SSIM, on every comparable image.
• Lossless. Smol always re-encodes. For pixel-identical output use ImageOptim, OxiPNG or Caesium’s lossless mode.
• JPEG and WebP against Squoosh. A tie, not a win: the same bytes for the same score.
• JPEG speed against ImageOptim. ImageOptim processed 24 JPEGs in about 0.9 seconds; Smol took about 7.1.
• Same format against ImageOptim. By SSIMULACRA2, JPEG to JPEG and PNG to PNG are ties, with ImageOptim’s JPEG 4% smaller at a score of 80, and Smol’s WebP was 4 to 6% larger than ImageOptim’s JPEG at the same score. By SSIM, ImageOptim’s JPEG was 1 to 8% smaller than Smol’s.

• Images. The 24 images of the Kodak Lossless True Color Image Suite, 768 × 512 pixels each, 15,394,305 bytes of PNG in total. They are public, so anyone can repeat this.
• Machine. Apple M2 Pro, 16 GB, macOS 27.0.
• Tools. Smol 1.0.35 driven through its MCP server; sips-316 (built into macOS); caesiumclt 1.5.0; squoosh.app in headless Chrome 154; Zipic 1.10.2 on its free tier, which allowed three images and no AVIF.
• Settings. Each encoder at four to ten quality settings per format, except Squoosh’s AVIF (default only). 1,335 output files in all.
• Scoring. Every output against its original PNG with SSIMULACRA2 (libjxl 0.11.2) and SSIM (11-tap Gaussian window, sigma 1.5, mean of the red, green and blue channels).
• Matching. For each image and encoder, the bytes needed to hit a target score are interpolated in log-bytes between the two nearest settings measured, then summed over the 24 images. If any image cannot reach the score, no total is reported.
• Timing. Wall-clock seconds for the whole folder, two or three runs each, all listed above.
• ImageOptim, 9 October 2026. ImageOptim 1.9.3 run from its command-line binary on copies of the files, on the same machine under macOS 27.0.1, at its defaults and with lossy mode on at JPEG quality 60, 70, 80 and 90 and PNG quality 60, 80 and 90. Smol 1.0.35 was run on the same inputs; its JPEG, WebP and AVIF curves from the PNG originals are the 5 October rows, made with the same version. 696 more output files, scored and matched the same way, over the images both tools reach at each score. ImageOptim’s times include launching the app; Smol’s run from the job being accepted, with the app already open.

• Twenty-four small film scans are not your photo library. Phone photos, screenshots and product shots on white will give different numbers.
• A quality metric is a model of what people see, not people. SSIMULACRA2 and SSIM disagree in places, which is why both are shown.
• One machine, one version of each tool, one day. Smol’s JPEG output changed between versions 1.0.34 and 1.0.35, and other tools change too.
• Squoosh’s AVIF was measured at one setting and Zipic on three images, so those comparisons are narrower than the rest.
• ImageOptim was tested four days after the other tools, and its JPEG inputs were made by us from the PNG originals. A camera’s own JPEGs will behave differently in detail.
• We make one of the products tested. The data and scripts below exist so the result does not depend on trusting us.

• results.json and results.csv: one row per scored file (image, tool, format, setting, bytes, SSIM, SSIMULACRA2), 1,335 rows.
• ssim_lib.py, harvest2.py and match2.py: the scoring and matching code.
• smol-run.mjs and squoosh.mjs: how Smol and Squoosh were driven. README.txt explains the rest.
• ImageOptim test: results.json and results.csv (696 rows, with a column for the input format), io-run.sh, smol-run2.mjs, harvest-io.py, match-io.py and a README.

For what the numbers mean when you are choosing a tool, including four compressors we assessed without measuring, read the best image compressors for Mac. Found a mistake? Tell us and we will correct it.

• ImageOptim alternative for Mac: when to switch, and when to keep ImageOptim.
• The best image compressors for Mac: what these numbers mean when you choose a tool.
• AVIF converter for Mac and compressing images for the web.
• Image file size and page speed, image optimization for web agencies and client delivery for photographers.