Apple reveals new AI architecture built around Google Gemini models
Points and comments are a snapshot, not live.
Apple unveils AI architecture using Google Gemini models for on-device and cloud processing.
Apple announced a redesigned Apple Intelligence platform built on Apple Foundation Models co-developed with Google using Gemini technologies. The system includes five models: two on-device variants (AFM Core and AFM Core Advanced for standard and multimodal tasks), three cloud-based models running on Apple's Private Cloud Compute infrastructure (AFM Cloud, AFM Cloud Image, and AFM Cloud Pro). Cloud Pro, the most capable model, runs on Nvidia GPUs in Google's cloud under Apple's privacy guarantees and offers Gemini frontier-level quality. New capabilities include realistic image creation, advanced photo editing, visual question answering, speech generation, and improved dictation. A system orchestrator coordinates features across platforms, tailoring responses based on active apps and user tasks. Apple emphasized on-device processing and Private Cloud Compute preserve privacy, with user data unavailable to Apple or third parties.
What commenters are saying
Commenters sought clarity on the collaboration's technical details: whether Apple uses Gemini behind custom prompts, fine-tunes it, or wraps it wholesale. Early analysis from Apple's keynote revealed five distinct models, with on-device variants "refined" using Gemini and Cloud Pro described as having "quality similar to Gemini frontier." Sentiment split between those concerned about Google involvement (despite data privacy assurances) and those reassured that Google does not access user data. Secondary discussion noted Gemini's hallucination issues in public releases and questioned whether Apple's deployment sidesteps these problems through different training or constraints.