Google has allowed Apple to distill Gemini models so they run directly on iPhone and Mac

Apple and Google: New Details on AI Collaboration

Last week Bloomberg journalist Mark Gurman reported what can be expected from Siri in iOS 27. At the same time, more concrete information emerged about an alliance between Apple and Google in artificial intelligence, including issues of control over Gemini family models.

What has changed in the partnership?
- Deeper than anticipated – the partnership turned out to be “deeper” than experts had expected.
- Greater freedom for Apple – the company now has substantially more opportunities to work with Google’s technologies than previously thought.
- Full access to Gemini – Apple can now use Gemini models in its own data centers and adapt them to its needs.

A source from The Information noted: “Apple has full access to the Gemini model in its data centers. This allows the company to create more compact versions that perform specific tasks or are so small they can run directly on Apple devices, speeding up request processing.”

How does it work?
The process of creating derivative models is called distillation.

- One “teacher” model transfers knowledge to more compact and optimized algorithms.
- Compact models run faster and require fewer resources, making them suitable for local operation on user devices, unlike large models that are typically run in the cloud.

Because Apple has full access to Gemini, future models can “learn” Gemini’s internal computations and mimic its behavior. This will improve training efficiency and allow the creation of smaller models that perform close to their “teachers” while consuming far less computational power.

What remains unknown?
- Apple’s goals – the process can be complex because Siri’s tasks do not always align with Gemini’s specialization.
- Proprietary AI models – the company continues to develop its own solutions, but specific objectives and timelines have not yet been disclosed.

What to expect at WWDC?
Apple plans to unveil a significant update to Siri in June at WWDC. It is expected that the voice assistant will be able to:

1. Remember past conversations – improve contextual support.
2. Support proactive features – for example, suggest leaving home earlier to avoid traffic and arrive on time at an airport or train station if the user plans a trip.

Thus, collaboration with Google opens new opportunities for Apple to develop Siri, making it more intelligent and convenient for users.