For me, the beauty of Beancount[0] is that it's just text files in Git. There's a web UI I use for generating reports, and a Python API with which I hacked together some import/export scripts, but 99% of my interactions with it are via Emacs[1] and Magit. GnuCash may use XML to store its data, but I'm pretty sure it's not designed around editing those XML files directly.
A ton of repetitive bookkeeping tasks become so much easier when you can copy and paste, or use keyboard macros or something like multiple-cursors[2], rather than have to click tens or hundreds of times in a GUI. Many years ago I used QuickBooks, and basic tasks like importing a bank statement took at least an order of magnitude longer than they do now.
Having my company's books in Git is also huge when it comes to auditing, concurrency, backups, and figuring out where things went wrong when accounts don't balance. As mentioned in another comment: `git diff` is a really powerful tool and it's awesome to be able to check out the books as they existed at a particular point in time. `git blame` is great for when you're trying to figure out where a number came from. Writing meaningful commit messages and comments keeps me sane when I try to remember a year later why something is recorded the way it is.
The biggest downside—or advantage, depending on how you look at it—is that there's no default or built-in chart of accounts, so you need a certain level of accounting acumen (or professional advice) to set things up at first. I'm pretty sure GnuCash aims to be plug-and-play, whereas Beancount is more akin to a programming library that you use to build an accounting system. I agree with the grandparent commenter, who said that text-based accounting is "the best and most flexible accounting experience I've ever had." But the cost of that flexibility is that there's a fair amount of DIY involved.
[0]: https://beancount.io/
[1]: https://github.com/beancount/beancount-mode
[2]: https://github.com/magnars/multiple-cursors.el
