Apple to Rely on Google’s 1.2 Trillion-Parameter AI Model for Siri Enhancement

Apple is strategically leveraging Google’s advanced capabilities to enhance the functionality of its reimagined Siri, with the introduction of a powerful custom Gemini AI model in the cloud.

Insights from Mark Gurman: Apple to Implement a Custom Gemini AI Model with 1.2 Trillion Parameters for Siri Overhaul

In recent reports by Bloomberg’s Mark Gurman, it has emerged that Apple is set to integrate a custom Gemini AI model, boasting an impressive 1.2 trillion parameters. This tailored model significantly surpasses the 1.5 billion-parameter proprietary AI that currently operates behind Siri in the cloud.

Prior to finalizing this unique approach, Apple evaluated other leading AI models including OpenAI’s ChatGPT and Anthropic’s Claude. Ultimately, the decision to adopt a customized version of Google’s Gemini was made to facilitate more complex user interactions, all within the boundaries of Apple’s Private Cloud Compute system, which emphasizes user privacy through the use of encrypted and stateless data.

Furthermore, as part of this agreement, Apple is expected to invest approximately $1 billion annually to access Google’s cutting-edge AI technology. This deal represents the latest chapter in the enduring commercial partnership between Apple and Google, wherein Google also compensates Apple $20 billion per year for default search engine rights in Safari and other Apple platforms.

Internally, Apple has designated the initiative to refine Siri with an internal codename, Glenwood. This project is being led by notable figures including Mike Rockwell, the architect behind the Vision Pro headset, and Craig Federighi, head of software engineering.

The Comprehensive Siri Revamp: Key Components

Apple’s upcoming redesign of Siri will feature three fundamental components:

Query Planner – This segment represents Siri’s operational tier, tasked with determining the optimal path to address a user’s request. This could involve conducting web searches, accessing personal data (like calendar or photo records), or interfacing with third-party applications through App Intents, a framework that enables Siri to provide in-app actions without user intervention.
Knowledge Search System – Siri will incorporate a comprehensive knowledge database designed to manage trivia inquiries independently, minimizing reliance on third-party AI or online search results.
Summarizer – This integral feature within Apple Intelligence enables Siri to utilize external AI models, such as ChatGPT, to condense various text or audio inputs, alongside functionalities like notification summaries and webpage insights.

In the context of Apple’s evolving technological landscape, the customized Google Gemini model is expected to primarily manage the Query Planner and Summarizer functionalities, while the Knowledge Search System will be powered by Apple’s proprietary on-device LLMs. Importantly, this arrangement is designed to maintain independent functionality without positioning Google’s search AI at the forefront of Apple’s ecosystem.

While this partnership is a critical step forward, Apple envisions developing its exclusive solution over time. For now, the Gemini model will play an essential role in launching the upgraded Siri experience with the release of iOS 26.4, which will introduce enhanced in-app actions, contextual awareness, and improved visual interactions for the voice assistant.

Source&Images