Apple Intelligence in China: Qwen Gets Siri, Writing Tools, and Visual Intelligence, Apple Keeps the Base Model
The dual-model architecture—a self-trained base plus a third-party model for knowledge-heavy entrances—gives teams a template for shipping AI in regulated markets, and the compliance timeline shows how to stage a release when filings gate every step.
Apple Intelligence reaches mainland China in two steps: a November 4 developer beta on iOS 27.3 and a December system update for all users. The first batch ships eight deterministic productivity features: notification summaries, live translation, writing tools, visual intelligence, photo cleanup, custom emoji, text-based Siri, and priority mail. The new-generation Siri is not part of this wave.
The structural change sits in the model routing. Alibaba's Qwen enters three system-level entrances—Siri, Writing Tools, and Visual Intelligence—where general world knowledge is required. Apple's self-trained China-specific model keeps the base and system actions, mirroring the global architecture that routes some requests to OpenAI's ChatGPT. A two-and-a-half-year wait came down to the compliance chain: approval in July, self-trained model news in August, Alibaba's model cleared on October 14, and the announcement on October 31.
Hardware and conditions gate the rollout. Supported devices start at iPhone 15 Pro, with some models using up to 8 GB of storage. Four conditions must all pass: the device is supported, iOS is 27.3 or newer, both device and Siri languages are English or Simplified Chinese, and free storage is at least 8 GB. The documentation includes a Python function that codifies this gating.
The launch gives teams a working example of feature-model decoupling, compliance treated as a milestone, and trust built on deterministic capabilities. The absent new Siri marks the next round.
Choosing eight deterministic, low-risk features for the first wave is a deliberate failure-mode control: familiar features with low per-error cost keep first impressions manageable.
Qwen's placement inside three primary interaction entrances marks the first time a Chinese large model holds a main interaction position in a foreign flagship product, which carries more weight than benchmark rankings.
The dual-model route—base self-controlled, entrances localized—serves as a reference for any overseas AI product entering China, because the compliance chain forces a two-layer design.