Three. Every concept comes back from the edit as one core video with three different openers, because the first seconds decide most of an ad's fate and no one can predict which opener wins. Testing three gives the idea itself a fair trial, separate from the luck of its first impression.
The hook-level readout then feeds everything: when an opener wins, it gets grafted onto other concepts, and when a concept fails with all three hooks, the idea is retired with confidence instead of a shrug. That distinction, idea versus opener, is where most in-house testing goes wrong.
