CSS Formatters and Minifiers Compared: Which Ones Can Eat Your String

There is one question that separates every tool in this list, and it is not "does it produce pretty output." It is: does the tool know what a string is?

A CSS minifier that does not parse CSS will happily eat content: "x/*y*/z" and hand you content:"xz". That is not a hypothetical edge case; it is what happens when "delete the comment markers" is a regex. Everything below is sorted by that line, not by size or speed.

Figures are npm downloads API pulls from 2026-10-12, 30-day windows at each end of 2024-10-01 → 2026-10-10. Last-publish dates read from the registry the same day. Install counts, not users, and not choices — a transitive dependency is credited as loudly as a picked one.

The table

Read the last-publish column before the change column

clean-css grew +56% and has not shipped since 2023-11-30. It is still installed 100.3M times a month. Whatever you think of that sentence, it is the most transferable fact on this page: a package can be the fourth most-installed CSS transformer in the ecosystem and be three years dead, and nothing in a package.json will warn you.

cssnano shipped on 2026-10-12 — the day of this pull — and grew the least of any actively maintained tool here. Slow growth on a maintained package is not the same signal as fast growth on an unmaintained one, and the two are easy to confuse if you only look at percentages.

Which is which

• Lightning CSS (lightningcss) — Rust parser with a WASM build, minifies and transforms, vendor-prefixes as a side effect. The +130× is mostly vendoring: bundlers and frameworks compile it in rather than exposing it, so a large share of that number is other people's build pipelines, not people choosing a CSS tool. Treat it as a ceiling on how much CSS parsing moved inside tools, not a measurement of adoption. The trends page has the check I ran against the Tailwind confound, and it did not fully resolve.
• PostCSS (postcss) — not a formatter, a platform: it parses CSS into an AST so your plugin can rewrite it. 1.342B installs in 30 days means it is infrastructure under everything else in the table.
• cssnano — the minifier that runs on PostCSS, so it gets a real syntax tree and can be configured to leave alone what it does not understand. Slower and larger than a Rust minifier; that is the price of the tree.
• clean-css — a standalone minifier with its own tokeniser. Unmaintained since 2023. Modern syntax (nested rules, @layer, color-mix(), container queries) postdates its last release, and "it minified my modern CSS into something else" is a bug class that only gets worse with time.
• Prettier — a code formatter that happens to handle CSS. It will not minify, will not optimise, and is genuinely good at making CSS readable in a commit. Different job from every other row here.
• stylelint — a linter. It reports; it does not rewrite. It is in this table because "which formatter" and "should anything touch this file at all" are the same conversation.
• This site (csstidy.net) — textual, deliberately. No parser, no install, no upload; your stylesheet never leaves the browser. That privacy guarantee and the string-eating limitation are the same design decision, not two separate facts.

The test you can run in ten seconds

Take any tool in this list and feed it this declaration:

.a::before { content: "x/*y*/z" }
.b::after  { content: "}" }

A real parser returns both strings intact. A textual minifier returns content:"xz" — it removed four characters from inside a string literal, and it will do so silently, in CI, on your production bundle. A textual beautifier has a worse version of this failure: the } inside the string truncates the block, which is reproduced against this site's own shipped code on the About page, including the fact that the minifier survives it and the tidier does not.

I have not run Lightning CSS or cssnano on those two lines in this session, so I am not claiming measured results for them — I am claiming the definition. A tool built on a syntax tree has the information required to get it right; a tool built on string replacement does not. Verify with the command above on the version you actually ship.

What the numbers cannot tell you

• Transitive vs chosen. postcss at 1.342B is mostly other tools' dependencies. It is the floor of the ecosystem, not a popular choice.
• Correctness. No cell above is a statement about edge cases. The 10-second test is, and it costs less than reading this page.
• Whether you need any of them. If you are editing one stylesheet by hand, a formatter you run once is not a dependency decision. If you are formatting in CI across a team, you are choosing a contract, and the right move is to pin one and stop arguing about it.