It's surprising that anyone still cares about the rar file format. lzma, as used in .7z, has superior compression, and neither are particularly fast so it's not about performance.
7-Zip is BSD licensed and has a native Windows UI.
Granted it doesn’t have compression advantage over 7z, but those flags and features look great when I want to create archives, generally better and more convenient than anything else I look at, but I usually end up going with plain old zip files since various utilities can scan and search through them, etc., a network effect win for the zip format. But it also underscores that the best compression ratio doesn’t count for that much for me and some other people
Not a scientific benchmark, but I think it underlines general point. If I want the best results I use a .tar.xz at insane compression levels, or more commonly a .tar.zstd if I want good decompression speed. The usecase for 7zip and WinRar is convenience, ease of use and windows-native file handling instead of the unix-focused .tar format. WinRar wins out on all three of those.
7zip's gui is a worse clone of WinRar's, archive creation has a fraction of the features, windows-specific file handling is an afterthought at best (winrar has handling for the archive flag, alternate data streams, file security, hard links, etc). And most important of all rar is built as an archive format. You get built-in recovery records, and hashes are stored as blake2 hashes instead of the frankly insufficient crc32 hashes 7zip uses.
I'd give 7zip points if it had a better (== more familiar) CLI, but they made the bizarre decision to copy winrar's cli too and make even worse documentation for it. The only things it has going for it are a linux UI and the open-source license
About lzma, I believe zst beats it on every metric. I resisted it for a while, because it's tainted by Facebook, but I can't argue with the results.
Performance, yes. But not compression. lzma compresses much smaller than zstd.
There was a particularly good benchmark comparison of the two which put them on a Pareto front of speed vs size from different levels over each on some different sources but I can't find that. ZSTD was ahead most of the time.