The best time, as per my knowledge and my own experience would be to use this feature in your first 24-48 hours post publishing your content. The main objective of this very feature is to help you to determine whether or not your current metadata or thumbnail is helping you in boosting the overall performance of your video or content. It helps you to understand how overall your content is doing and helps you in the decision-making process whether you need to work on your Title or your Thumbnail to make the analytics look better.
Of course, you will see a bit here and there in your video performance but in end, it will give you the results you're looking for. There will be moments when you might notice that your CTR or overall video impressions or reach has been reduced when there's A test going on and as compared to B, you might see a bit fall in your views but that's what the A/B testing stands for, to compare the results and to help determine which results you should move forward with.
The reason I'm suggesting that it's best to do such A/B testing post 24 hours is that, when you post a video, it takes the first 3 hours to overall show you the entire performance of the video including the CTR, AVD and number of views and I personally, let the video stay for at least 24 hours to see how well it's doing before trying out the tests to find out what could be improved.
---
In regards to how long you should run such a test, I'd say the first 14-28 days is enough to get an idea what are the things you should go ahead with, average video performance and its future is determined by how well the video did it in its first 3 days performance as YouTube will mark "video" as "New" tag and will try to push the video to a wider audience as much as possible, and in such 72 hours, YouTube will check how your audience is reacting to your video, what's your average watch duration, how're people responding to the impressions of your video on their home page or search results and everything. So for the start, you can go ahead with 14 days of A/B testing, which will give you great comparison results on which test did great and which didn't, accordingly, you can determine or decide, which thumbnail or metadata you should be using for your video.
In the end, it's still a personal preference, If you're a legend user then make sure you play with the A/B testing tool as much as you can so you can experience different results and experiment with it, such experiences will help you to determine different theories and you can use such knowledge for your upcoming video testings. You wouldn't know until you have tried it so let's give it a try and see!
