Technical report detailing the foundation models powering Apple Intelligence. ~3B on-device model and larger server models running on Apple's Private Cloud Compute. Uses a proprietary mixture-of-experts architecture. Trained with adapter-based fine-tuning for specific tasks (summarization, proofreading, Siri).

The 2025 update added multilingual support (16 languages), multimodal capabilities, and improved tool use and reasoning. Distinctive for its on-device + private cloud split, processing data locally where possible and using dedicated Apple Silicon servers when not. Proprietary.

Paper

arXiv: 2407.21075

on-devicemultimodalmoe