Revise Blog - Building a post-AI word processor

A drag race between Revise, Word, and Google Docs

Software that you use every day should be fast!

Artur Sapek, Founder of Revise··3 min read·

I've always been impatient with slow software. Apps should launch quickly, and get out of the way.

In building Revise I've really tried living by this rule. With how good computers and edge networks are these days, a word processor really has no excuse. It should open files and give you a cursor instantly. I'm pretty happy with where the product is at performance-wise, so I created a benchmark to showcase it compared to Word 365 and Google Docs.

I ran two tests: opening an existing document from your account, and importing a .docx file as a new document. Each test ran against three different test documents of varying lengths: 2 pages, 30 pages, and 140 pages.

In both tests Revise is consistently faster to initialize and become interactive than both Google Docs and Word 365. Word and Docs sometimes put text on the screen quickly, but kept loading for longer before the rest of the UI was actually ready. The full results are below.

Opening the 30-page report with an empty cache in Revise, Word 365 Online, and Google Docs. One run of each, recorded on October 19, 2026. Timers count from the click.

Opening an existing document

This first test measured how long it takes to open a document already in your account. How long does each product take to first show the document text, and then to become interactive? Measured by frame-by-frame analysis of screen captures.

Google Docs sometimes shows document text a little faster, but Revise generally wins when it comes to fully loading. Docs and Word take longer to fully load their formatting ribbon and other UI controls before the editor is ready for use.

Importing a .docx

The other big one: how long does it take to import a .docx file and load it in the editor? Revise has its own custom DOCX parser which runs entirely client-side, and I think this chart really shows what difference that makes. Revise is the clear winner with all three test documents.

Revise imported the 2- and 30-page documents in under a second, Word in 2.7 to 2.8 seconds, and Google Docs in 4 to 5. On the 140-page manuscript, Word showed text about as soon as Revise, and slightly sooner with a warm cache.

Why Revise is faster

I think there are a few technical reasons why Revise is so much faster. It's served as a static app on an edge network (CloudFront), so time-to-first-byte is very quick. It has a custom DOCX parser which runs client-side, so it relies less on server processing. It's also generally a lean application and uses lazy-loading effectively, downloading about a tenth of the bytes Word does and a sixth of what Google Docs does, while offering comparable DOCX compatibility and feature parity.

Most of the download is the app itself, not the document: going from the 2-page memo to the 140-page manuscript added about 1 MB in Revise and Word, and 0.2 MB in Docs. Revise also made far fewer requests, about 120 against about 290 for Docs and 870 for Word.

How this test was run

  • Setup: one Chrome 155 window (1440 by 900) on an Apple M4 Pro Mac, on a modern WiFi network. Tests ran in Chrome using Playwright.
  • Documents: a 2-page memo, a 30-page report (The Wealth of Nations, with tables and footnotes), and a 140-page manuscript (Pride and Prejudice).
  • Text on screen is the first frame where text recognition finds the document's first sentence. Interactive is when the app's toolbar also settles (within 1% of its final look), since each app shows a greyed-out or partial toolbar until its editor is ready. It is a visual measure; the script did not type. In Revise, text and toolbar appeared in the same frame in every run.
  • Runs: each case ran at least five times warm and five times cold (caches cleared, still signed in).
  • Download size was measured separately, on October 19: one load of each document's link in a blank tab with caches cleared, counting every response and WebSocket message received in the first 15 seconds.
  • Data: runs.jsonl, bytes.jsonl, excluded.jsonl (every left-out run and why), and meta.json. Scripts and documents: speed-benchmark.zip.

The Revise engine

You can use the Revise document engine and DOCX parser in your own software! As of a few weeks ago they are available on NPM!

npm i revise

Use this to build document processing into your own applications, or give it to AI agents as a CLI - it will be able to work with Word documents more quickly.

The Revise API also offers a simple hosted version of our engine which you can use to run our agent harness against your own documents, on our servers.

Q&A

Why is Revise faster than Google Docs and Word?
Mostly because it loads less. With an empty cache Revise downloaded 2 to 3 MB to open a document, against about 16 MB for Google Docs and 23 to 24 MB for Word, and most of those bytes are the app rather than the document. Revise also opens a document inside the page that is already loaded instead of loading a new one, and it keeps documents you have opened before on your device, so a warm open can show it before the server responds.
Is Google Docs faster than Word 365 Online?
It depends on the task. Google Docs finished opening the 2- and 30-page documents sooner than Word, and the 140-page one later. Word imported .docx files faster in every case.
How was this benchmark run?
Each run was a screen recording of one Chrome window (1440 by 900) on an Apple M4 Pro Mac over WiFi, driven by Playwright and signed in to each app. The open test clicked the document on each app's home screen, and the import test uploaded a .docx through each app's own import or upload button. Timers start at the click. Text on screen is the first frame where text recognition finds the document's first sentence. Interactive is the first frame after which the toolbar stays within 1% of its final look, because each app shows a greyed-out or partial toolbar until its editor is ready. Each case ran at least five times with a warm cache and five times cold (caches cleared, still signed in). Download sizes were measured separately, from one load of each document per app. The scripts, the documents, and every run, including the ones left out and why, are linked in the method section.

Recommended reading