Apple is working on one of the biggest upgrades to its voice assistant since its launch. According to a Bloomberg report, the company is close to signing a contract with Google worth around one billion US dollars per year. The goal is to integrate a version of Google's Gemini model into the next generation of Siri. The price isn't the only interesting factor; the architecture of the new technology is particularly crucial, as it will determine how powerful, efficient, and secure the new Siri will actually be.
For years, Siri has been considered nice, but technologically outdated. While voice models like ChatGPT, Gemini, and Claude communicate more and more naturally, Siri has mostly remained limited to simple commands. That's about to change. The planned upgrade, powered by Google's Gemini model, aims to make Siri not only smarter, but also faster and more privacy-friendly. Crucial to this is its technical architecture – a system likely based on the so-called "mixture of experts" principle.
The partnership between Apple and Google
According to Bloomberg, Apple and Google are currently negotiating a close collaboration. The deal would obligate Google to provide a specially adapted Gemini model with 1.2 trillion parameters. This model would run on Apple's private cloud compute servers, ensuring that Google has no access to user data. This is crucial from a data privacy perspective because all requests and calculations remain within Apple's infrastructure. Google would thus provide the technology, while Apple retains control over the data and its processing.
1.2 trillion parameters – what does that actually mean?
The planned version of Siri is expected to operate with 1.2 trillion parameters. Parameters are the building blocks of an AI model—they control how it understands and contextualizes language and generates responses. A model of this size is impressive. For comparison, Apple's current cloud-based model is said to have around 150 billion parameters. The new version would therefore be many times larger. However, a direct comparison with other large AI systems like GPT-5, Gemini 2.5 Pro, or Claude Sonnet 4.5 is difficult. Many major AI labs no longer publish their exact parameter counts. Estimates range from under a trillion to several trillion parameters. No one knows for sure. Nevertheless, with 1.2 trillion parameters, Siri would clearly be among the largest AI models currently available.
Why architecture is crucial
The sheer size of a model says little about its efficiency. What matters is its internal structure. According to several indications, Apple is likely using a Mixture-of-Experts (MoE) architecture – a technique employed in many current high-performance models. In this approach, the model consists of several specialized subnetworks, known as "experts." For each input, only the experts relevant to the specific task are activated. This ensures the system remains efficient, even with a very large number of parameters overall.
How Mixture of Experts works
In a Model of Excellence (MoE) system, the entire model is not activated simultaneously. Instead, it reacts selectively: For each input, the system selects only a few of the available experts, usually between two and four out of several dozen. Assuming the Siri model operates with 32 experts, only about 75 to 150 billion parameters would be active per request, even though the entire system comprises 1.2 trillion. This drastically reduces computational costs while simultaneously increasing speed. In theory, Siri can thus utilize the power of a massive model without fully consuming its resources.
Apple's approach to data protection
A key element of the new Siri remains data privacy. Although Google provides the model, the processing runs exclusively on Apple's private cloud compute servers. This means that Google has no access to requests or user data. Apple can thus continue to deliver on its most important promise: personal data remains private. The use of Gemini doesn't change this; it simply adds a more powerful AI layer to Apple's existing infrastructure.
Open questions regarding performance
It remains unclear how the Gemini-based Siri will compare to other models when it launches next year. The sheer size of the model is no guarantee of superior performance. Factors such as the quality of the training dataset, the optimization of the inference processes, and the integration with iOS also play a role. Apple has been using its own cloud AI systems based on MoE technology for some time now. The current model, with around 150 billion parameters, serves as the foundation. The new, Gemini-powered model would therefore represent a further development – larger, more efficient, and presumably significantly more context-aware.
Does Siri have the potential to compete with ChatGPT and Gemini?
Whether the new Siri can truly compete with modern AI systems like ChatGPT, Gemini 2.5, or Claude 4.5 remains to be seen. Crucially, it will depend on how well Apple adapts the Gemini model to its devices, software, and privacy policies. However, with its planned architecture and considerable number of parameters, Siri has the best prerequisites to become significantly more intelligent and versatile than before. If Apple masters the interplay of performance, efficiency, and security, Siri could become a voice assistant by 2026 that not only executes commands but also understands and intelligently guides users through complex tasks.
Siri is on the verge of the biggest technological leap in its history
The collaboration between Apple and Google marks a significant step in the evolution of voice assistants. With 1.2 trillion parameters, a mix of expert technology, and a strict approach to data privacy, a completely new generation of Siri is emerging. It should work faster, more precisely, and more flexibly than ever before. At the same time, user privacy remains protected – a key factor that could clearly differentiate Siri from many competing systems. If the integration proceeds as planned, Siri will become one of the most powerful and secure AI assistants of its time. The best products for you: Our Amazon Storefront offers a wide selection of accessories, including those for HomeKit. (Image: Apple)
- iOS 26.2: When the update will be released & which features are coming
- Apple is making progress: Will Siri finally become intelligent in 2026?
- iPad mini with OLED: Everything about price, performance and launch
- Apple in November 2025: All developments at a glance
- Apple launches major smart home offensive with new devices
- Will Apple's first budget MacBook arrive this year?
- Apple TV 4K Update: All rumors about the new model at a glance
- iPhone 18 Pro: First rumors about design and technology
- AirPods Max 2: All the rumors about the design, chip, and market launch
- Apple 2026 in transition: Five innovations will shape the year
- Apple on the verge of change: How the leadership team is changing
- Apple Glasses combine fashion, technology and AI in everyday life
- Apple Vision Air: Why the headset may never be released
- Apple announces new immersive content for Apple Vision Pro
- iPad mini 8: All rumors and information at a glance
- Severance Season 3: All information about the start, cast and filming
- Apple is preparing more product innovations: Apple TV & more
