The question was primarily about background capability.

Thank you, I will file a feedback ticket. Just a note: If Core AI is the way for developers to deploy custom (non-Apple Intelligence) models, it would be consistent that Core AI supports non-Apple Intelligence devices, at least to the same degree as Core ML, so there are concrete reasons to migrate to Core AI. Without support for A14, A15 and A16, the value of AOT compilation drops significantly in 2026.

Related: What about security patches? Do they also invalidate the cache?

Related: Are there any specific caveats when using Core ML and Core AI at the same time for multiple models in a single process?

Great question