Hands-on with Apple Intelligence on a China-market iPhone: Siri can call Qwen, translation and object removal have highlights

Just now, with the iOS 27.3 developer beta update, Apple Intelligence—which we've waited more than two years for—has finally opened for testing to users in China!

From its first appearance at WWDC 2024 to now actually being installed on China-market devices, Apple's AI pie, two years in the making, has finally been served up to us.

More importantly, Apple has found a solid local partner for the China version this time: Qwen.

The China version of Apple Intelligence is powered by a foundation built jointly on Apple's in-house models and Alibaba's Qwen. Features like photo cleanup and basic polishing run directly on-device; but once it encounters complex tasks, the system calls on Qwen's cloud capabilities through Siri, Writing Tools, and Visual Intelligence.

As for which iPhones can use it, how the China version differs from the overseas version, and issues like Apple accounts and regional restrictions, we have already laid this out in detail before—interested readers can head over to read "Just now, China-market Apple Intelligence was officially announced, but full-power Siri AI will still have to wait."

First, search for and download the official "Apple Developer" app in the App Store, and sign in to your Apple ID in the account section; after a short wait, open the system's "Settings - General - Software Update," and the latest iOS 27 Developer Beta will show up under "Beta Updates."

After all, this is the first developer beta firmware, so some bugs are inevitably present in the system. If you only have one main phone you use heavily every day, it's advisable to wait until the official release in December before getting on board safely.

Are Siri, photo editing, and translation any good?

This system update's package is huge—a full 12GB—so be sure to leave enough storage space on your phone before upgrading.

And because many China-version AI capabilities rely on on-device model processing, new data indexing needs to be established, so right after getting started there are some stability issues—for example, natural language search in Photos, Memory movies, and lock screen notification summaries all need to wait for indexing to complete, and it may take one to two days before they work normally.

Next, let's put a few key features to the test:

The all-new Siri

Leading the charge, of course, is Siri. Gone is the "artificial idiot" that could only set alarms and check the weather, and would toss you a few web links if you asked anything slightly deep. The Siri with its new marquee-light skin finally supports double-tapping the bar at the bottom of the screen to interact by typing directly.

In our tests, if you just ask simple everyday questions or give straightforward basic setup commands, it can handle them directly. But as soon as you take a slight detour, Siri still gets confused.

As for cramming two tasks into one sentence, or deeper complex system orchestration, it is still powerless—in terms of system control, at heart it's still the same familiar Siri as before.

So what about more complex, lengthy tasks? We threw it a request to write a guide: "I want to go to Jiuzhaigou for four days next month. Help me put together a travel guide."

Faced with a question beyond its own capabilities, the China-version Siri pops up a window asking whether you agree to call Qwen, and notes that you need to sign in to a Qwen account (if you choose to refuse the call here, Siri will simply give up answering outright.)

After tapping agree to call, it generates an answer in about 7 or 8 seconds. Judging by generation speed alone, it's indeed not slow, but looking closely at the content, it's a bit embarrassing. The so-called four-day guide it gives is extremely brief in both accommodation suggestions and attraction arrangements, with very low information density—at best it can only be considered a very preliminary outline preview, and the actual result is far inferior to asking directly in the Qwen app.

Photo cleanup

Compared with the still somewhat wooden Siri, the "Clean Up" feature in the Photos app is much more immediately noticeable; this capability calls on Apple's on-device model.

In our tests, Apple Intelligence's cleanup feature performs quite commendably.

Specifically, if the subject to be erased in the frame is relatively far from the camera, has clear outlines, and the surrounding background is fairly clean, the cleanup effect is very neat and tidy, with minimal loss of image quality—a single photo's size only drops by about 0.2MB.

But once the subject you ask to remove is more complex—such as the large areas of grass and cobblestone pavement in our test image—it's more prone to failure. After cleanup, the painted-over area leaves behind a blurry patch of very low resolution that clashes with the originally clear image quality around it.

Live translation

This capability likewise relies on the on-device model; on first use you need to download the language pack for the corresponding translation language, each about 2GB.

Apple has tucked this capability into apps like Messages, Phone, and FaceTime, and even when connected to AirPods, it can achieve "translate while you chat."

In our tests, the live translation feature performs surprisingly well in everyday spoken conversation—it responds quickly, with less than 1 second of latency, and its accuracy is decent too, enough to support smooth communication with foreigners.

Writing Tools

Anywhere you can input text, select the text and you can call up Writing Tools to rewrite, proofread, and summarize—Writing Tools is definitely the feature office workers are most likely to use. But for the China version, what we care about more is whether this AI writing tool imported from overseas will be "unable to adapt" in a Chinese context?

For example, take a rather blunt deadline-chasing message: "You said you'd deliver the proposal on Friday, and now it's Monday—why hasn't it been sent over yet? Please handle it as soon as possible." When we tried rewriting it with the writing tool, even after deliberately switching to the gentlest "Friendly" mode, the rewritten sentences still felt stiff and formulaic. And the additional features like summaries and bullet points are all on par with most ChatBots.

However, when it comes to ease of access, Apple has leveraged its system-level advantage. In our tests, anywhere there's a text input field—whether it's the built-in Notes app or third-party apps like WeChat, Feishu, and DingTalk—it can be easily invoked, which is quite convenient.

Visual Intelligence

This time Apple has also brought recognition capabilities into the real world—Visual Intelligence, which can be activated by long-pressing the "Camera Control" button, with two main interaction entry points: "Ask" and "Search."

The "Ask" on the left taps into Alibaba Qwen's capabilities; the idea is to let you ask questions directly about the object in front of you, for example by taking a photo of a skincare product and asking about its key ingredients and uses.

The "Search" on the right is closer to the familiar "reverse image search," using Baidu's visual model by default, and can quickly retrieve similar images that can be found on Baidu.

But during testing we discovered a pleasant surprise: besides Baidu image search, many apps can also provide similar functionality by integrating the corresponding API interfaces.

For example, with Dewu installed on the phone, it appeared in the list of selectable search services and retrieved similar items for sale on Dewu; even a niche chord-query app I had installed on my phone, Guitar Wiz, also appeared in the list of selectable search services and could be used to look up chords.

It's not hard to infer that Apple has most likely opened up the corresponding API interfaces to third-party apps—as long as developers have adapted them and the user has the corresponding app installed on their phone, its search capabilities can be invoked directly through Visual Intelligence. We're very much looking forward to seeing how apps like Taobao and Xiaohongshu perform once they're integrated.

Humorous.

How far is Apple from true intelligence?

After going through the first batch of features of the China-region Apple Intelligence one by one, friends who normally follow Apple should have a good sense by now—put plainly, this is the "initial version" of Apple Intelligence that first rolled out overseas in 2024, not Apple's latest phone AI solution.

In fact, in the overseas beta of iOS 27.3, Siri AI already supports Traditional Chinese, and its capabilities have been further upgraded, increasingly showing signs of maturity.

By comparison, the just-updated China-region Apple Intelligence seems somewhat lacking in "intelligence."

But this is still a milestone achievement.

After more than two years of waiting, Chinese users' iPhones have finally filled in this long-missing system-level AI piece of the puzzle. What's more, for the vast majority of ordinary people, no one really expected a omnipotent sci-fi butler to suddenly pop out of their phone—they just hoped Siri would stop being clueless like before and could use useful features like live translation and object removal.

If we dig deeper, has this iPhone undergone any "qualitative change" because of Apple Intelligence?

Obviously not.

This current batch of features is essentially still an "AI +" empowerment approach—Photos plus AI becomes photo object removal; the input field plus AI becomes text polishing. Here, AI is merely an "external plugin" assisting the original apps, not the phone's underlying logic.

What we truly look forward to is AI taking over underlying interactions: when you say to your phone, "Add the restaurant address my friend just sent to tonight's schedule, and check how long it takes to get there from the office," it can recognize the friend, find the address, write it into the schedule, and handle everything for you from start to finish.

In fact, the new version of Siri AI Apple launched this year focuses on screen awareness and contextual understanding, and has certain personal context management capabilities.

Even so, over the past period, the vast majority of users have given rather lukewarm feedback on Apple Intelligence and even the new Siri AI:

The experience is far from as amazing as in the promotional videos.

Ifanr believes the key to this gap still lies in long-range action capabilities and personal memory management.

The phone is perhaps everyone's most intimate electronic device—it can not only remember your life trajectory in the background over the long term—distinguishing who your "friends" are, finding where your "company" is—but also cross permissions to search chat records and mobilize the calendar and maps.

Currently, Apple is working hard to connect Siri with the various native apps on the iPhone so that data can flow and scheduling becomes smoother.

But our digital lives are still highly tied to third-party apps like WeChat, Meituan, and Feishu. No matter how seamlessly Apple connects its own Notes and Calendar, as long as it can't get inside the walls of third-party apps, its understanding of you will be fragmented.

Frankly speaking, even if we can soon use the China-region version of the new Siri AI smoothly, it won't have an earth-shaking impact on our daily lives in the short term. For Apple, how to build a good personal context so that the AI in the phone can better serve people remains something that will take a long time to develop.

Over the past two years, we've seen too many dazzling "order coffee with one sentence" demos at various launch events, but once you actually use a phone as your daily driver for a few months, you'll find that the wow factor in a demo is still a hundred thousand miles away from the stability and reliability needed for everyday use.

Take Apple as an example again: despite just reporting the best fiscal-year earnings in its history, it still faces enormous challenges—

With a massive base of more than 2.5 billion active devices and 1.5 billion paid subscriptions, the commercial value of Apple AI isn't just about getting consumers to upgrade their devices; it's also a reason for users to spend more money.

On this earnings call, Apple executives practically laid this commercial strategy out on the table. CFO Kevan Parekh explicitly mentioned that AI will create new monetization opportunities through its own services and partners; new CEO John Ternus even outright revealed that in the future, features relying on cloud-based large models, such as Siri AI and advanced intelligent photo editing, will have a basic usage quota, and heavy users will need to pay for more quota.

This also means that iCloud+, which in the past mainly used cloud storage space to attract users to pay, will likely transform into Apple's subscription gateway for selling "cloud computing power and advanced intelligence." Apple clearly doesn't intend to measure the value of this long-term investment solely by the performance of this generation of products.

But no matter how attractive the business blueprint is, that's still a long-term story Apple is telling Wall Street; coming back to the iPhone in our own hands, it's better to keep our expectations in check for now.

As for when we'll actually get to use that true AI Phone, whether it's Apple or the phone industry as a whole, we may just have to wait a little longer.

This article is from the WeChat public account "Ifanr" (ID: ifanr), author: Fang Junming, editor: Xiao Qinpeng, published by 36Kr with authorization.