Smol vs Dinky: We Fact-Checked Their Comparison Table

Most of the time, and we mean that literally. Here is the shape of it.

You want to spend nothing. Dinky is MIT. Not freemium, not a trial, not a Pro tier. Its own FAQ says “No trial, no watermarks, no ‘pro’ tier.” If the honest requirement is “make my screenshots smaller and stop asking me for money,” there is no argument to have. Download it and stop reading.

Disk space actually matters to you. 35 MB against 403 MB is not a rounding error, it is roughly eleven times the footprint, and on a 256 GB MacBook Air that is a real consideration. Our number is larger because the bundle carries the upscaling models, the document and spreadsheet toolchain, and encoders for file types Dinky does not touch. That is an explanation, not a defense. If you only ever compress images, you are paying disk rent on capabilities you will never open.

You want to read the source. Smol is closed. Dinky is not, and its README names every bundled engine with its licence: cwebp (BSD), avifenc (BSD), oxipng (MIT), qpdf (Apache 2.0). If “I can audit what touches my files” is a requirement rather than a preference, that settles it.

You need OCR on scans. Dinky runs Apple’s Vision OCR on scan-like PDFs before the flatten or preserve pass, so a scanned contract comes out smaller and searchable. Smol has no OCR. A scanned page goes in as pixels and comes out as smaller pixels. For anyone who archives paperwork, that single row is worth more than everything else on this page.

You live in a shell. Dinky’s CLI is a real CLI: dinky compress-image, dinky compress-video, dinky compress-pdf, dinky ocr, with --json output and a dinky serve loopback endpoint behind a bearer token. You build it from the repo yourself. Smol has no shell binary. Our automation surface is an MCP server, which is a different shape and does not help you inside a Makefile.

You want ALAC or AIFF. Dinky cross-converts audio through afconvert with bundled LAME for MP3, and covers ALAC and AIFF. Smol writes AAC, M4A, MP3, WAV, FLAC, OGG and Opus. We tested AIFF input on 26 September 2026 and it was rejected outright.

Four rows. Scraped verbatim from dinkyfiles.com/compare/smol/ on 26 September 2026. Each one is checkable in about a minute, and we have said how.

1. “$29 one-time (7-day trial)”. There is no trial. We are putting this first because it is the one that flatters us and it is still wrong. Smol has no downloadable trial build. There was one; there is not one now. The /trial URL on smolmac.com is a permanent redirect to the buy page, and the checkout has a single path: pay $29, get a licence key. Anyone who read that row, downloaded expecting seven free days, and hit a paywall was misled in our direction. We would rather lose that click than keep it.

2. Audio: “—”. Smol compresses and converts audio. The compression targets are AAC, M4A, MP3, WAV, FLAC, OGG and Opus. The conversion targets are M4A, MP3, WAV, FLAC and OGG. Bitrate is adjustable from 16 to 512 kbps and sample rate from 8 to 192 kHz.

Since a capability list is only an assertion, here is a measurement anyone with a Mac can repeat. Take /System/Library/Sounds/Blow.aiff, which ships with macOS, and convert it to WAV so both apps can read it:

afconvert -f WAVE -d LEI16@48000 Blow.aiff Blow.wav

That produces a 273,248-byte file, 48 kHz 16-bit stereo, 1.40 seconds long. Dropped into Smol at 128 kbps on an Apple M2 Pro, 16 GB, macOS 27.0 (build 26A428), it came back as Blow_compressed.m4a at 23,078 bytes in 849 ms, a 91.6% reduction, verified as AAC by afinfo. The same file converted to FLAC produced 44,155 bytes in 73 ms. Take the ratios with the usual pinch of salt, because uncompressed PCM compresses spectacularly and a 1.4-second alert sound is not an album. The point is only that the column is not empty.

One genuine gap turned up while testing. Smol rejected the original Blow.aiff with “This file couldn’t be processed.” We do not read AIFF, and we do not write ALAC. Dinky does both. That belongs in the audio row too, and it belongs on Dinky’s side of it.

3. Watch Folder: “—”. Smol has watched folders. They are exposed in the app and over the MCP server as four operations: list, add, update and remove. On 26 September 2026 we added /tmp/smol-watch-test through the MCP server, got back a live configuration with the id muhzix6478ada2, and removed it again. A folder can carry its own preset, so a watched ~/Dropbox/Inbox can compress to different settings than a watched ~/Desktop/Screenshots.

4. CLI / automation: “—”. Half right, and worth splitting carefully. If the row means “a shell binary you can put in a script,” Dinky is correct and we are not going to pretend otherwise. Smol has no smol command.

What Smol has instead is a Model Context Protocol server, installable in one click for Claude Code, Codex and Google Antigravity. An agent calls smol_compress, smol_convert, smol_upscale, smol_strip_metadata and the watched-folder and preset tools directly, with jobs it can poll and cancel. Every measurement on this page was produced that way. There is also a Finder Quick Action, which reported installed: true on this machine, and system-wide clipboard compression. Three automation surfaces is not “—”.

Dinky, to be fair, has its own answer here and it is a good one: dinky serve is a token-authenticated local HTTP endpoint the README describes as being for “scripts and AI-agent workflows.” Different protocol, same instinct.

Two smaller things while we are here. The row reading “Apple Silicon only · macOS 13+” has it backwards: Smol ships separate native builds for Apple Silicon (macOS 13+) and 64-bit Intel (macOS 14+), while Dinky’s own README says the 2.x DMG and Homebrew cask “are not for Intel-based Macs.” And “AI upscaling: 2× / 4×” misses the middle step, which is 3×.

Corrections are cheap. Here is the part that costs us something.

Dinky is 35 MB and we measured Smol at 403 MB on disk (du -sk /Applications/Smol.app, version 1.0.34). Dinky’s README says “The whole app is 35 MB, still appropriately dinky,” and their comparison table listed ours as “Not published.” Now it is published. Eleven times the disk for a compression utility is a real cost, and anyone who finds that absurd is entitled to.

It is also free, under MIT, with 513 stars and 65 releases on github.com/heyderekj/dinky since the repository opened on 16 April 2026. That release cadence is not typical of a free Mac utility. The most recent commit on main as of 26 September 2026 was 261d8d1b, pushed the previous evening. Compare that to the state of Squoosh, which no maintainer has touched since August 2024, or Optimage, last released in November 2021 and still Intel-only. Dinky is the opposite problem to write about: there is nothing stale to point at.

The feature-by-feature wins are specific. OCR on scan-like PDFs, which we do not have at all. URL drop, paste a link up to 500 MB and it downloads, compresses and cleans up the temp copy. Smart quality that auto-detects photo versus graphic per image. A skip threshold so files that would save less than 2%, 5% or 10% are left alone. Session history with paths and before/after sizes. An Apple Shortcuts action. A Homebrew cask, which is how a lot of developers would rather install anything.

Derek Castelli, who writes it, describes himself on the repo as a freelance web designer working in Webflow and Figma who got tired of compressing site images by hand. That is exactly the kind of person who builds a good small tool, and it shows in the defaults.

Scope, mostly, and two specific outputs Dinky cannot produce.

Dinky will not give you a JPEG back. Its image outputs are WebP, AVIF, lossless PNG and HEIC. That is a deliberate and defensible design choice, because those four are where the real savings live, and its README argues the point well. It is also a hard stop if the destination is a platform that only accepts JPEG, a print supplier, a government upload portal, or a CMS from 2014. Smol writes JPEG, PNG, WebP, AVIF, HEIC and GIF on compression, and adds TIFF, BMP and PDF as conversion targets.

Neither will it touch a document. Smol converts Word to PDF, DOCX, ODT, RTF, TXT and HTML; spreadsheets to PDF, XLSX, ODS, CSV and HTML; presentations to PDF, PPTX and ODP; tables to CSV, TSV, JSON, NDJSON, Markdown, XLSX and Numbers; and fonts to TTF, OTF, WOFF and WOFF2. Dinky is a compressor for four media types and says so plainly. If your actual pile of files is a client handover with a PPTX, three XLSX exports and a folder of PNGs, that is one app instead of two.

Then the edges. On-device AI upscaling at 2×, 3× and 4×. A neural mode that hits an exact byte target anywhere from 4 KB to 4096 KB, where Dinky’s equivalent binary-searches quality to an MB target on images only. Video out to MOV, WebM and MKV as well as MP4. An Intel build. macOS 13 support, which matters if you are on a 2020 Mac that Sequoia never reached. Developer ID signing and Apple notarization, so nothing sends you into System Settings on first launch. An interface in 50 languages against 13.

And the MCP server, which is the one that will either mean nothing to you or be the whole reason. If you use Claude Code or Codex, Smol appears as a set of tools the agent can call. “Compress every PNG in public/ to WebP under 200 KB and rerun the build” becomes a step the agent executes rather than a chore you remember afterwards. The full tool surface and the guardrails are on the MCP page.

Read down the left column and stop at the first row that describes you.

There is no row that says “both,” but installing Dinky first and buying Smol only if you hit one of its walls is a perfectly sensible order of operations, and it is what we would suggest to a friend.

If you are comparing more broadly, Smolr vs Smol covers the other free open-source Mac compressor with a confusingly similar name, ImageOptim alternatives covers the free lossless route, and Squoosh alternatives covers the browser option. Smol itself is a one-time $29 purchase covering one Mac, with no subscription and no per-file limits.