Apple Enables Frontier Models Running Locally On iPhone

Apple will be able to run pruned frontier models on iPhones within three years, according to a post shared on X referencing Baker; those on-device models could process roughly 30–60 tokens per second and operate without cloud connectivity. If realized, this local, private approach could sideline cloud-dependent model builders, reduce AI data-center demand, and position Apple as the dominant distributor of everyday AI functionality.