It's different from A/B testing in that the experiment is explicitly designed to cause harm to half of the participants.
Presumably most A/B testing would be designed to figure out which choice performs better on a set of metrics. But going in, there is little evidence to point to one or the other, and the "harm" caused would simply be in user experience. In this experiment, the researchers had a prior theory about which choice would cause harm, and the harm is emotional and psychological.
All that aside, if this was purely internal research at Facebook, it would still likely be unethical but probably nothing out of the ordinary. The fundamental different is that this is being presented as scientific research. It's published in PNAS. It involve three co-authors from various universities. There are standards, both legal and ethical, that must be followed when engaging in scientific research, and the concern is that such standards were perhaps not followed.
Manipulation and even inducing harm may be widespread throughout the advertising industry, but that's advertising, not science.