It doesn't make any sense to look at Metal scores for the GTX 1060 as there is no Mac shipping with that GPU. This is most likely some weird hack.
On Windows, GTX 1060 scores around 35K, irrelevant of the API.
Anyway, M1 doesn't perform too well in Geekbench Compute, my guess it's because of the slower memory bandwidth. It doesn't do too well in LuxMark either. At the same time, many real-world GPU compute applications are sensitive to data transfer latency and M1 will perform much better in those. For example, some initial reports of M1's TensorFlow performance show that it rivals the GTX 1080.