Apple overhauls AI with new architecture built on Google Gemini

Apple today revealed a significant revamp of its Apple Intelligence platform, built on a new architecture that integrates foundation models co-developed with Google using Gemini technologies. The models are adapted to run both on-device and on Apple's Private Cloud Compute servers, enabling a 'huge upgrade' in understanding, reasoning, and multimodal capabilities. This includes realistic image creation, advanced photo editing, and visual question answering. A new system orchestrator coordinates features securely across platforms, tailoring responses based on the active app and user task, delivering what Apple calls truly system-wide intelligence. Apple positioned this as a privacy-focused alternative to competitors, reiterating that user data is only used for immediate requests and is not accessible to Apple or third parties, with outside experts able to verify these guarantees at any time.

Higher-power versions of the model will be available on select devices, offering additional capabilities like speech generation, improved dictation, and stronger natural language understanding, though Apple did not specify eligibility. The announcement highlights a deepening collaboration between Apple and Google, leveraging Google's Gemini family to enhance Apple's AI offerings while maintaining Apple's privacy-first stance. This architecture aims to differentiate Apple in the AI race by combining state-of-the-art model capabilities with its existing infrastructure and strict data-handling policies.