Siri AI rollout: advanced voice expressivity requires 12GB of unified memory

Following the rollout of its next-generation Apple Intelligence and Siri AI, Apple published which devices get which capabilities. Baseline Siri AI features run across existing supported hardware. The more advanced on-device features, custom voice expressivity and local generative capabilities, require at least 12GB of unified memory paired with M3 or M4-class Apple silicon or the iPhone 18 Pro. 9to5Mac compiled the full device list, and it leaves many recent iPhones and Macs with the reduced feature set.

The technical reason is memory capacity and bandwidth. Local generative voice models must hold weights and runtime state in RAM while producing speech in real time, and Apple's privacy-first approach pushes as much work on-device as possible. A 12GB floor suggests models large enough that quantization alone cannot squeeze them onto 8GB devices without unacceptable quality or latency.

Competitively, Apple is taking the opposite approach from cloud-first assistants. Google's Gemini 3.8 Flash TTS clones voices from 30-second samples in the cloud, and Meta's Muse and Microsoft's Copilot run heavy work server-side. Apple's bet is that privacy and latency justify hardware gating. The gating doubles as an upgrade driver for iPhone 18 Pro and newer Macs. Third-party entrants are also targeting Apple's messaging surface, including szn, an iMessage agent from a former Siri engineer.

The skeptical take is that feature fragmentation frustrates users and developers, and that memory-based gating conveniently aligns with hardware sales. Watch reviews of Siri AI quality on baseline devices, and whether Apple offloads the gated features to Private Cloud Compute for older hardware.