That was unexpected; I liked the LANDR master better.
Of course, a sample of one doesn't prove anything. As panup says "it's hit or miss depending on the mix".
I had similar results using Har-Bal in full-automatic mode - one in ten songs came out better than my own version. But that was 10 years ago. At that time I'd not previously done any mastering. I had mid-grade speakers in a small room with no acoustic treatment. The last time I revisited Har-Bal was about 5 years ago, and in ten out of ten tests it made the mixes worse, not better.