Apple Intelligence in China Opens for Testing with iOS 27.3 Developer Beta, Siri Integrates Qwen
Apple Intelligence in China has opened for testing with the iOS 27.3 developer beta, with the official release expected to roll out in December. The feature is jointly supported by Apple's self-developed on-device model and Alibaba's Qwen: simple tasks are processed locally, while complex queries invoke Qwen's cloud capabilities after obtaining consent via a pop-up, and the Qwen integration is off by default. The first batch launches basic features already released overseas in 2024, such as photo cleanup, writing polish, real-time translation, and notification summaries, and does not include the next-generation Siri AI showcased at WWDC 2026; it supports only iPhone 15 Pro / Pro Max, the iPhone 16 series, and later models, and first-time activation requires downloading about 7GB of on-device model content.
Why read this
AI score 70
The first batch of Apple Intelligence in China only fills in individual features; the invocation and authorization mechanism of Siri's deep integration with Qwen is key to understanding this rollout.
Topics
Body · AI translation
Article body EM_StockImg_Start
EM_StockImg_End
Apple Intelligence for mainland China opened for testing on Wednesday with the iOS 27.3 developer beta. In this version, Siri is deeply integrated with Alibaba's Qwen, and the official release is expected to roll out in December.
A reporter for the Star Market Daily has learned that Apple Intelligence for mainland China is jointly supported by Apple's self-developed on-device model and Alibaba's Qwen. Simple tasks are processed locally; for complex Q&A or creation, such as in-depth Siri questions and content generated by writing tools, the cloud capabilities of Qwen are invoked after obtaining user consent via a pop-up.
The first batch to launch consists of the basic features already released overseas in 2024, including photo eraser, writing polish, real-time translation, notification summaries, and so on, and does not include the new-generation deeply understanding Siri AI demonstrated at WWDC 2026. The latter is the "new-generation Siri" rebuilt on Google Gemini, featuring in-depth cross-app operations and long-context understanding, and the mainland China version has still not launched it.
In terms of device models, Apple Intelligence for mainland China only supports iPhone 15 Pro / Pro Max, the iPhone 16 series and later models (including Air/Duo). iPad and Mac are currently not supported. Enabling it for the first time requires downloading about 7GB of on-device model support content, and some features, such as natural-language search in Photos, require waiting 1-2 days for the index to be built before they can be used normally.
Based on hands-on experience from multiple developers, after waking Siri there is a rainbow marquee effect, and typed communication is supported. However, some developers report that in hands-on testing it is not very capable of handling complex system scheduling such as "two tasks in one sentence"; complex questions need to rely on Qwen, and after the jump, the answer quality shows no obvious difference from that in the Qwen app directly.
On October 31, 2026, Apple and Alibaba officially announced their cooperation; Qwen is integrated into Siri, writing tools and visual intelligence, and in December it will be officially pushed to Chinese users with iOS 27.3.
A reporter for the Star Market Daily noted that this integration mechanism is similar to the way the overseas version of iOS connects to ChatGPT, but control is explicitly returned to the user. Qwen integration is off by default, and users need to actively choose to enable it, either through a Qwen account or "Sign in with Apple"; the system seeks consent before sending questions, files or images to Qwen; Alibaba promises that users' requests and generated replies will not be used to train the Qwen model.
It is worth mentioning that after this update, the system already supports real-time translation for AirPods. One tester said, "English translation is very accurate and fast, while Japanese and Korean perform average."
A consumer electronics analyst told a reporter for the Star Market Daily that Apple Intelligence for mainland China mainly covers the first-generation overseas features, only with the third-party domestic support switched to Qwen. "It basically meets expectations, and there is not much surprise. The official version in December is expected to be improved."
In October, StepFun officially released its first AI agent phone, the STEPX Neo, and simultaneously launched the agent-native operating system Step AOS and the personal agent Amoo. With a single sentence from the user, "Help me book a hotel and plan an itinerary," the system links ecosystem services such as Ctrip and Alipay, all without opening separate apps.
ZTE and ByteDance jointly advanced the second-generation Doubao phone, the Nubia NaviX Ultra, which was also released in September. It follows the route of system-level integration with the Doubao phone assistant, with the focus on moving from "passive response" to "recognizing intent and proactive service."
Honor's Magic9 series, meanwhile, is equipped with the system-level Agent Harness commercial architecture; the YOYO agent achieves a comprehensive task accuracy rate of 91.8%, can complete long-horizon tasks of over a hundred steps, and has jointly built five vertical-domain models with Alibaba, pushing "can understand" toward "can execute stably."
In a horizontal comparison, this launch of the Apple Intelligence version is mainly about filling in AI features: single-point capabilities such as photo eraser, writing proofreading and notification summaries are stuffed into the system one by one, while the Siri AI that truly handles multi-step task orchestration has not entered the first batch, and cross-app execution capability is still absent.
(Source: Cailian Press) Body middle embedded ad