Will my privacy be compromised with the new Siri using Google's Gemini?

According to official statements, no. Apple uses a "distillation" technique, leveraging Google's large model to train its own compact model, ensuring that user interaction data with Siri is not sent to Google, effectively borrowing the model's capabilities without surrendering data.

When can Taiwanese iPhone users access the new Siri, and how well will it support Traditional Chinese?

The new Siri will be available with the iOS 26.4 update, but advanced features are typically tied to newer devices and systems. The actual performance of Traditional Chinese support will require real-world testing, and users should have realistic expectations, as Apple's localization efforts often take time to refine.

Doesn't Apple have its own AI capabilities? Why is it partnering with rival Google?

Apple's in-house models are lagging behind in the large language model space, making it challenging to support a sufficiently intelligent Siri. After evaluating OpenAI and Anthropic, Apple chose to "rent" Google's strongest model to address its shortcomings while using distillation to safeguard user privacy and relationships.

As an app developer, will this impact me?

Yes. Apple is replacing the old SiriKit with App Intents, and developers who want to integrate their apps into the new Siri voice experience will need to plan for migration. Securing an early position in the voice entry point for 1.5 billion devices has strategic value.

Apple Pays Google $1 Billion Annually for Gemini-Powered Siri: What This Deal Means

Apple is spending approximately $1 billion per year to lease a customized 1.2 billion-parameter Gemini model from Google to drive the new Siri. Why has the typically self-sufficient Apple chosen to outsource its AI brain this time, and what does this mean for the 1.5 billion Taiwanese device users?

How long has it been since you last seriously asked Siri to do something for you? I asked my friends around me, and most of them replied, "Setting alarms, playing music, nothing more." This voice assistant, which has been criticized for several years, will get a brand-new "brain" in 2026 - and this "brain" is not developed by Apple itself, but rather rented from its competitor, Google.

For a company known for doing everything itself and keeping everything under its control, this decision is news in itself.

Event Background

Apple and Google announced their cooperation on January 12 this year, with Apple obtaining the right to use a customized Gemini model for Siri and Apple Intelligence. This deal is estimated to be worth around $1 billion per year. At the WWDC 2026 in June, the new Siri powered by Gemini was officially unveiled and is expected to be pushed to around 1.5 billion daily active devices worldwide through iOS 26.4.

The dramatic aspect of this event lies in the fact that Apple's AI progress has been criticized for being slow in recent years, and its own model has not been able to support a smart enough Siri. After evaluating the solutions offered by OpenAI and Anthropic, Apple chose Google, citing that "Google's technology provides the most capable foundation." In other words, this once-proud player has acknowledged that it needs to borrow someone else's "brain" in the field of large models.

Key Points

Customized 1.2 billion-parameter Gemini model: This model is tailored by Google for Apple, with a scale approximately eight times that of Apple's current cloud model (around 1.5 billion parameters). It adopts a hybrid expert (MoE) architecture, specializing in optimizing tasks such as summarization, planning, and natural language understanding, which are essential for Siri.
Around $1 billion per year: This is the level of licensing fees Apple pays to Google.
Protecting privacy with "distillation": This is a crucial design aspect. Apple engineers use Google's "teacher model" to train a more compact Apple base model. When users interact with Siri, their data will not be sent to Google, equivalent to borrowing a brain without sharing data.
Developers need to rewrite their code: At WWDC, Apple also announced the replacement of the old SiriKit with App Intents, requiring developers to migrate their apps to be compatible with the new Siri.

Market Impact Analysis

For Taiwanese users: If you are an iPhone user, the most direct change is that Siri may finally "understand human language and handle complex tasks." In the past, Siri was limited to single instructions, but with the support of Gemini, the new version is expected to process multi-step, contextual requests. However, there are two practical reminders: first, the actual performance of traditional Chinese needs to be verified after launch, as Apple's localization in Chinese often requires time to refine; second, these advanced features are usually tied to newer models and systems, so older iPhones may not be compatible.

For enterprise applications: This deal amplifies a signal - even Apple chooses to "rent" top-level models instead of building its own. For Taiwanese enterprises, this is a kind of endorsement: you don't have to train models yourself; it's practical to connect to the most powerful existing models and spend resources on your applications and data. The focus shifts from "should I do AI myself" to "how can I smartly use others' AI."

For developers: The replacement of SiriKit with App Intents is a clear action signal. Taiwanese developers who want their apps to be compatible with the new Siri and integrated into this AI experience need to start researching migration. This is both a threshold and an opportunity - getting your app connected to the voice entry of 1.5 billion devices has considerable value.

Future Development Trends

The most intriguing aspect of this deal is that it makes two competitors interdependent. Google gains $1 billion per year and the badge of "even Apple chooses me"; Apple supplements its weakest link at a relatively affordable cost and preserves its privacy trump card through distillation. While this is a win-win situation in the short term, whether Apple will slowly nurture its own model and replace Google in the long term is worth watching - after all, "borrowing a brain" has never been the end goal for Apple; self-control is.

Another trend is that "models as infrastructure" are becoming increasingly evident. Top-tier large models are turning into basic services like electricity and cloud computing, provided by a few companies and used by others for a fee. This has a profound impact on the power structure of the entire industry.

TheAI Academy Summary and Commentary

Many people interpret this news as "Apple admits defeat," which I think is an oversimplification. The truly smart aspect is that Apple separates "whose model is the strongest" from "who controls user data and experience" - the brain can be rented, but the entrance, devices, privacy, and ultimately the user relationship are firmly in Apple's hands. Renting a brain is far more cost-effective than handing over users.

The essence of Apple's move is not "using Gemini," but "using Gemini without letting Google touch my users" - this is the consistent calculation of this company.

Practical advice for Taiwanese readers: don't rush to expect Siri to transform overnight; the actual performance of traditional Chinese and device limitations need to be verified after launch. However, if you are a developer, migrating to App Intents is worth doing now; if you are a general user, take this as a signal that Siri is worth giving another chance, and expect improvements with a rational and practical mindset.

Data Sources

(This article is compiled based on publicly available information and is subject to Apple and Google's official announcements. Functionality, launch schedules, and device support are subject to the latest official information.)