Why are similarity scores lower with text-embedding-3-small?

FWIW I was disappointed with text-embedding-3-small and found it less useful. I have stuck with text-embedding-ada-002

I’m slightly surprised we’ve not seen more embedding models.