Apple overhauls AI architecture with Google Gemini foundation models
The tech giant describes the collaboration as a "deep" partnership that unlocks significant upgrades to its Apple Intelligence platform, while maintaining strict privacy controls through Private Cloud Compute.
Apple has announced a significant restructuring of its Apple Intelligence platform, introducing a new architecture centred on foundation models co-developed with Google using technologies from the Gemini family. The company described the partnership as "deep," stating it unlocks what it termed a "huge upgrade" for the platform, bringing state-of-the-art understanding and reasoning capabilities alongside multimodal support for image understanding and generation.
At the core of this revised system is a new orchestrator designed to coordinate Apple Intelligence features securely across all Apple platforms. This central component allows the system to tailor responses based on the active application and the user's current task, enabling what Apple describes as truly system-wide intelligence. The architecture supports both on-device processing and server-side execution via Apple’s existing Private Cloud Compute infrastructure, ensuring that data handling remains within the company’s controlled environment.
The update introduces a two-tiered structure for on-device intelligence, with certain devices receiving a higher-power version of the model. This enhanced tier offers additional capabilities, including speech generation, improved dictation accuracy, and stronger natural language understanding. However, Apple did not specify which hardware qualifies for this superior model, nor did it detail the specific chip and memory requirements, leaving the exact scope of device eligibility unclear.
Upgraded capabilities enabled by the new models include realistic image creation, advanced photo editing, and visual question answering. Apple positioned this approach in contrast to competitors it characterised as "racing forward" without regard for users. The company reiterated its commitment to privacy, stating that user data is used solely to execute immediate requests and is not accessible to Apple or third parties. To bolster trust, Apple added that outside experts can verify these privacy guarantees at any time.
This architectural shift follows the initial integration of Gemini models in iOS 27 and the Siri overhaul unveiled at WWDC 2026. The rebranded Siri AI is available as a standalone app with enhanced visual intelligence and cross-app context awareness. While iOS 27 expands device compatibility to all iPhone models from the iPhone 11 onwards, the new two-tiered model structure suggests a divergence in feature sets, with less capable devices receiving a reduced offering that lacks the new voice options and advanced capabilities reserved for the higher-power tier.
