Oddly enough, I just posted on this in a semi-related topic. I’ve offered some recommendations in that post.
As others have posted, there is no such thing as “best size”. It all comes down to what the application calls for.
At the end of the day, you will have to experiment but here are some tips I can share.
First, let me make a broad statement based on my many failed experiments. 
- Forget about adhesive mounted magnets. Even the strongest adhesive tape will fail over time especially if exposed to temp extremes and/or chemicals. You will have better results if you learn how to design cavities inside your model and embed them. You will also more likely love the resulting appearance and you will never have to worry about the magnet ripping off. There are many posts and YouTube video explaining how to embed hardware in prints such as nuts and magnets.
- You can find out for yourself but avoid ferro magnets like the plague. They are only good for gluing paper to a magnet or maybe holding something to a refrigerator whereby you’re only expecting to hold one piece of paper. Again, experiment and see for yourself if all you are doing is creating refrigerator magnets.
- Rare earth magnets are what you want to use if you are going to embed inside of a model. They are expensive by a long short over ferro magnets but do the job better than anything else. Try both ferro and rare earth and judge for yourself.
Considerations:
- Is this going to be a magnet that will have to be attached to a metal object or to another mating magnet?
- Embedded magnets that are concealed inside the print will work really well. That’s my favorite because they look great; it’s almost magical when you have an ordinary-looking model that can stick to something with no visible magnet. For that, bar magnets are best.
- If attaching a model to metal and desiring the concealed approach mentioned previously, then the question is: how thick is the intervening model wall? If you use 3mm thick magnets, you’ll be forced to make your model wall very thin if you want to go for the concealed look. My experience is that if you are doing metal-to-magnet attachment whereby the model is expected to stick on a metal surface, you’ll be forced to either make the wall thickness very thin or, as I recommend, go with a 5mm thickness magnet. Again, make test models to experiment.
- If you’re making, let’s say, two models that are going to stick together, such as a box with a magnetic lid, then you can easily use two opposing magnets, one on each side of the print. This is exactly what Bambu does with the nozzle cover. In that case, you don’t need thicker magnets; two opposing magnets work very well and round magnets are very suitable to this purpose. I might also recommend saving on CAD time and using the cut model feature in the slicer in combination with dowels which will give you a perfect press fit. (let me know if there is interest and I can post a tutorial)
- Round magnets are a great non-embedded alternative if you want to use a zero-thickness wall. In other words, the magnet will be exposed, similar to the magnets inside the Bambu lab nozzle cover. However, the trick here is to press-fit them into the hole and make sure the hole is slightly smaller than the magnet diameter so they don’t fall out. Again, as mentioned in the previous bullet, this can be more easily achieved with the cut tool and dowel feature.
And the last point, of course, is don’t forget that magnets will only stick to objects with a high iron content. Stainless steel will tend to be non-magnetic depending on the mix of Chromium, Nickle and Molybdenum content. So, you will want to test it against the surface you want to stick to. Some modern stainless steel refrigerators, for example, are designed specifically to not allow refrigerator magnets. I guess that’s to ensure a “clean” and “uncluttered” look. Clearly, those folks haven’t been in my kitchen. 