Mar 26, 2026 10:35:00

Apple may be able to create smaller in-device AI models using Google's Gemini.

Apple has reportedly

signed a multi-year agreement with Google to use Gemini, and will incorporate it into 'Apple Intelligence,' the AI system found in iPhones and Macs. New reports indicate that Apple will have full access to Gemini and be able to customize the model for Siri and other AI features.

Apple Can 'Distill' Google's Big Gemini Model — The Information
https://www.theinformation.com/newsletters/ai-agenda/apple-can-distill-googles-big-gemini-model

Apple Can Create Smaller On-Device AI Models From Google's Gemini - MacRumors
https://www.macrumors.com/2026/03/25/apple-google-gemini-distill-models/

Apple is rumored to be adopting Google's Gemini model for its smarter chatbot version of Siri in iOS 27, which is scheduled for release sometime in 2026.

Apple may be paying Google 150 billion yen annually to use a custom Gemini model for its new Siri - GIGAZINE

By integrating Gemini into Siri, it will be able to perform complex tasks that Gemini and other chatbots can do, such as answering questions, summarizing information, scanning and understanding uploaded documents, telling stories, providing emotional support, and completing tasks like booking travel.

According to the technology media outlet The Information, Google has given Apple 'full access' to its Gemini models in its data centers, which Apple can use to create smaller models for specific tasks and optimize existing models. In other words, Apple will be able to create models designed to run on devices without requiring an internet connection.

The Information reports that Apple is running a series of tasks on its main Gemini model and obtaining high-quality results along with an overview of the inference process. Apple can then use the answers and inference information from Gemini to train smaller, less expensive models. This process allows the smaller models to learn the internal computations used by Gemini, enabling the creation of more efficient models that perform at a similar level to Gemini but with less computational power.

However, while Apple can edit Gemini as needed to make it respond to queries in the way they want, they seem to be facing the problem that Gemini is tailored for chatbots and coding apps, and therefore doesn't necessarily meet Apple's needs.