I typically play two player Splendor, so I recorded myself playing against a single computer opponent to see how well I'd do. I do feel like the bots here play play pretty sloppy. Maybe they are better on the official Splendor app? Apologies is Vimeo is not your preferred video watching app, but I didn't want to needlessly clutter my YouTube or Twitch Channel.
https://vimeo.com/249109529My strategy is very similar to the one Awaclus demonstrated, and is largely inspired by it, except I don't tend to commit and reserve a card turn one. If I do, it would be for a really vital Level O card. Typically, my first couple of moves involve taking gems of colours matching ones that useful Level O cards require more than one of. This is obvious, but if a card requires one more gem of a colour you currently have, you can pick up that gem the turn before you buy it assuming you know the gem will be in supply by the time you want it. When a card requires multiple of the same gem, you need to collect that gem colour in advance.
My goal is generally to aim for ultimately getting a couple of Level OOO card, but I usually buy a Level O and Level OO card that ramp up to it first. Around the time I am building towards a second Level OOO card, I will look to reserve Level OO or Level OOO cards that will get me past the 15 point mark. Alternatively, I will reserve a Level OOO card but hold off on buying it and instead buy a chain of same-colour Level OO that require gems of their own colour. For example, both the 2-point 5G-cost Green card and the 3-point 6G-cost Green are on the board.
I also recorded a three player game against two bots, where I basically do the same thing.
https://vimeo.com/249103012So yeah, in short I am less averse to Level O cards as Awaclus.
Edit: Oh yeah, and denial reservations are more relevant in two player games, so there is a small added advantage to not holding three cards at a time in two player games.