Apple has their own models under the hood I believe. I remember from like a year or two ago they had an open line called "ELM" (Efficient Language Model), but I'm not sure if that's what they're actually using.

I am excited to see what the benchmarks look like though, once it's live.