Apple unveils the Apple Intelligence on-device foundation model
Apple Intelligence on-device foundation model unveiled — about 3 billion parameters, 3.7 bpw on average, TTFT of about 0.6 ms/token and 30 tokens per second on iPhone 15 Pro. Adapters loaded dynamically
Why it matters
The first large-scale case of OS-level on-device LLM integration
Source — Primary sources first; secondary reports are labelled “Press”
Officialmachinelearning.apple.comhttps://machinelearning.apple.com/research/introducing-apple-foundation-modelsAdded in Week 35, 2026 · See that week