Apple responded to the AI model controversy following WWDC 2026, emphasizing that Apple Foundation Models (AFM) are not “repackaged Gemini” models, but entirely independently developed. Although distillation and training processes leveraged Google’s Gemini technology, the final product and user interactions involve no Google code whatsoever. The AFM system comprises five customized models: the on-device model AFM Core, AFM Core Advanced (native multimodal, sparse architecture), the cloud models AFM Cloud, AFM Cloud Image (image generation and editing), and AFM Cloud Pro (agent tools)—all optimized for Apple Silicon and trained on proprietary data. Apple also confirmed that its cloud models run on Google Cloud servers and NVIDIA GPUs, while maintaining Private Cloud Compute certification to ensure user privacy.
Article author, source: chinaz
After the 2026 Worldwide Developers Conference (WWDC), Apple executives responded to media inquiries regarding recent speculation that its new AI model was a "repackaged version of Gemini." Apple clarified that its Apple Foundation Models (AFM) are not a simple replication of Google’s Gemini technology, but are entirely自主研发 and independently developed by Apple.
Apple noted that while its AFM model did leverage Google's Gemini technology during distillation and training, the final delivered product is entirely based on Apple's own code, technology, and data infrastructure. This means that users interacting with these models will not directly access Google's code, and the Gemini agent cannot access Google.
Currently, the Apple Foundation Models suite comprises five primary models: AFM Core, an on-device model focused on fundamental AI tasks; AFM Core Advanced, which features native multimodal capabilities and a sparse architecture to enable more complex AI functions locally. Additionally, the cloud-based AFM Cloud handles high-load requests, AFM Cloud Image specializes in image generation and editing, and AFM Cloud Pro is designed for agent tools and the most demanding workloads. Each model is customized for Apple Silicon and trained on proprietary data, with optimization via distillation using Gemini models.
Apple also emphasized that, despite using Google Cloud servers and NVIDIA GPU resources, it maintains its Private Cloud Compute certification to ensure user data security and privacy protection.