New Siri with Gemini: What Changes at WWDC 2026

The new Siri that Apple presented at WWDC 2026 is the biggest overhaul of the assistant in over a decade — and it no longer runs solely on Apple's own technology. Behind the more natural responses is a 1.2 trillion parameter Google Gemini model. In this guide, we separate confirmed facts from rumors and show what changes in practice.

TL;DR

The new Siri was unveiled at WWDC 2026 on June 8 as a conversational, multimodal assistant capable of performing tasks within apps.

It now runs on a custom Google Gemini model with 1.2 trillion parameters — about eight times larger than Apple's current cloud model.

According to Bloomberg, Apple will pay around $1 billion per year to Google for access to the model.

Features will roll out gradually throughout 2026, likely alongside iOS 27.

Sensitive processing remains on Private Cloud Compute, with an option to automatically delete history.

What Apple Announced About the New Siri

Apple used WWDC 2026 on June 8 to showcase a Siri practically rebuilt from scratch. The promise is to leave behind the command-by-command assistant and deliver something closer to a modern chatbot: continuous conversation, context memory, and the ability to chain multiple steps in a single request.

In the demo, the new Siri appeared in two forms. The first is the usual integration, embedded in the system and activated by voice or the side button. The second is a standalone app with a chat interface, putting Siri in direct competition with ChatGPT and Claude on the iPhone itself.

The most attention-grabbing point wasn't the interface, but the engine behind it. For the first time, Apple publicly admits that part of Siri's intelligence comes from an external model — and that changes the game for both users and developers in the ecosystem.

Siri Now Runs on a 1.2 Trillion Parameter Gemini Model

The heart of the overhaul is the partnership with Google. According to Bloomberg, Apple has signed a deal to use a custom Google Gemini model with 1.2 trillion parameters as the basis for Siri's cloud features. The estimated contract value is approximately $1 billion per year.

To put the leap in perspective: Apple's previous cloud model had about 150 billion parameters. The new model is about eight times larger and uses a mixture-of-experts (MoE) architecture, optimized for tasks like summarization, planning, and natural language understanding. In practice, it's the difference between an assistant that understands commands and one that understands intentions.

Before choosing Gemini, Apple also evaluated models from OpenAI and Anthropic. The decision to go with Google is a major shift in the company's foundation model strategy — the most significant since the launch of Apple Intelligence in 2025. It's worth noting that Gemini also starred in several recent Google announcements; we covered this movement in detail in our post on Google Gemini news at I/O 2026.

App Intents: Siri That Acts Within Apps

The most useful feature of the new Siri in daily life will likely be App Intents. Instead of just responding, Siri can now execute actions inside apps — even chaining steps across different apps.

A Concrete Example

Apple itself gave the example: Siri reads the weather app, notices it will rain, and automatically creates a reminder for you to grab a coat before your next appointment. That's three apps talking — Weather, Reminders, and Calendar — without you having to open any of them.

Why This Matters for Developers

App Intents transforms an app into a set of actions that Siri can call. For companies with their own app, this means part of the user journey can happen outside the traditional interface, via voice or text command. Those who expose their intents well gain relevance; those who don't become invisible to the assistant.

This is similar to what we've seen with assistants on other platforms and reinforces a trend: the interface of the future is increasingly conversational and task-oriented, not screen-oriented. In practice, this requires cataloging your app's actions and edge-case testing — because when Siri gets the intent wrong, it's your brand that takes the blame with the user, not Apple.

Onscreen Awareness and Personal Context

Two features underpin the feeling that the new Siri finally "understands" what you want: onscreen awareness and personal context.

Onscreen awareness allows Siri to interpret what's currently on the screen. You can ask for an action on the visible content without having to describe everything again — something like "reply to this message confirming the time."

Personal context goes further: Siri uses information from your device — emails, messages, photos, habits — to personalize responses. This is what allows it to know which "next appointment" you're referring to without you explaining.

Add to that the multimodal foundation: the new Siri processes text, audio, images, and video. There's even a camera mode where you point the device at an object and talk about what you're seeing. This is the natural evolution of features Apple began designing in iOS 26, with Liquid Glass and Apple Intelligence.

Privacy: Private Cloud Compute and Conversation Deletion

Using a Google model raises the obvious question: will my data go to Google? Apple's answer is no — at least in the design presented.

The company states that Siri's cloud features run within Private Cloud Compute, the infrastructure created to process sensitive requests without data being accessible to Apple or third parties. The Gemini model would operate in this isolated environment, not on Google's regular servers.

Additionally, Siri has gained retention controls similar to other chatbots. You can configure automatic deletion of conversations in three ranges: 30 days, 1 year, or never (keep indefinitely). For those handling sensitive data, this granular control is welcome — and necessary, given the type of information a personal-context assistant accumulates.

An honest observation: privacy here depends on actual implementation, not just marketing promises. It's worth following the independent audits that usually come out after launch.

Integrations: Safari, Messages, and Home

The overhaul doesn't stop at Siri. Apple showed Apple Intelligence going deeper into native apps:

Safari: automatic tab organization by topic and creation of browser extensions via text command.
Messages: action suggestions based on conversation content.
Home: analysis of security camera clips and generation of descriptions of what happened.
Camera: the new Siri mode, to interact with what's being captured.

One limitation already signaled: image generation may have daily restrictions due to computational cost. It's a reminder that generative AI at scale has a price — and even Apple can't escape that bill.

When It Arrives and on Which Devices

Here the ground gets slipperier, because part of the information is still rumor. What is known with more certainty:

Item	What Is Known
Announcement	WWDC 2026, June 8, 2026
System version	Likely to accompany iOS 27; there were rumors of an early release in iOS 26.4
Arrival window	Features rolling out gradually throughout 2026
Engine	Google Gemini model with 1.2 trillion parameters, on Private Cloud Compute
Languages	No official confirmation about Brazilian Portuguese at launch

Regarding hardware, Apple hasn't detailed the exact requirements yet. Since cloud features go through Private Cloud Compute, they likely depend on the same devices compatible with Apple Intelligence — which tends to favor newer models, as discussed in our overview of the iPhone 18 with A20 Pro chip.

What Still Concerns

Not everything is good news. Reports indicate that Siri's internal development was troubled, with successive delays — the revamped version was expected earlier and slipped to 2026.

The central concern of engineers, according to these reports, is accuracy in high-risk scenarios. An assistant that acts within apps needs to be almost always correct when dealing with sensitive categories like health or banking apps. A failed coat reminder is annoying; a wrong bank transfer is serious.

There's also the risk of dependence on an external supplier. Basing part of Siri on a Google model creates a relationship that could change with contract renewals, pricing, or regulatory disputes between the two giants. Apple has already signaled that it intends to return to using its own models in the future, indicating that the Gemini deal may be a bridge, not a destination.

Therefore, it's reasonable to expect a cautious launch, with features released gradually and prioritized markets. For Brazil, the recommendation is to calibrate expectations: a global announcement does not mean immediate availability in Portuguese.

What This Means for Brazilian Companies

Beyond the hype, there's a practical takeaway. The new Siri confirms that the voice assistant is no longer a bonus but has become a real interface layer — capable of reading, deciding, and acting.

For those with a digital product, three fronts deserve attention now:

Expose actions via App Intents. If you have your own app, map out tasks that make sense by voice and prepare for them to be called outside your interface.
Rethink content and customer service. Assistants that summarize and respond change how people search for information. Clear, structured content tends to be more "quotable" by these AIs.
Treat privacy as a differentiator. Apple's model puts privacy at the center. Companies that follow the same standard gain trust.

In the end, Apple's move follows a race already underway across all platforms — from Android to Google to corporate agents. If your operation doesn't yet have an AI strategy for customer service and processes, this is the sign that the topic has moved from "future" to "now."

Conclusion

The new Siri presented at WWDC 2026 is both a technical turnaround and a strategic admission: Apple needed Google to deliver the assistant it had been promising for years. With a 1.2 trillion parameter Gemini model, App Intents, and personal context, Siri finally seems modern — but success will depend on accuracy and a well-executed arrival in Portuguese.

If you want to understand how to apply conversational AI to your business — rather than just watch the announcement — it's worth talking to those who implement it in practice. It's exactly this kind of bridge between technology and results that guides our work at Agathas Web.