Neither. Nowadays it is only one part of a large scoring system using a bunch of device and user fingerprinting, and it's primary use seems to be tarpitting bots who score low on the other metrics (as well as any users unfortunate or privacy-conscious enough to be misidentified as a bot).

I'm pretty sure that's sufficiently shown by the fact that captcha v3 doesn't even have any inputs.

Yeah, its actually for their dataset.

Whichever saves/makes them more money.