It's a decent enough way to filter out obviously bad candidates without eating up your time (which is a big deal for small teams). I've used Codility for interviewing and being interviewed before and thought it was fine.
I would only reject a candidate if they got say less than 50% and the problems chosen weren't too difficult. If a company is going to be expecting 90% pass rates for super hard problems the rest of their interview process is probably faulty as well; that's not the fault of Codility.
I didn't say it was a fault with codility. Some companies, like your's, use it in a sensible way. But I think there are many companies which use this tool the wrong way. Anything except the easy level takes more time for someone who doesn't practice these kind of challenges every day. Moreover correct algorithm only scores 50%. 100% score can only be achieved upon perfect optimizing, which is very hard to do within 30 minutes. And your can't know which edge cases tests are failing until the solution is submitted after which its not possible to edit/correct it.
I'd say it's better for identifying really unsuitable candidates than it is for identifying who are the best ones.
codility for code challenges once in a while, yes. if its used just to see how you approach a problem and try to solve it, its good as it have nice code replays for the interviewer. But some companies want you to solve some random hard codility puzzle in 30 minutes or else you are no good