Apple Unveils Siri AI and Foundation Models Framework at WWDC 2026

Apple used WWDC 2026 to reset its AI trajectory, announcing Siri AI — a complete ground-up rebuild of its voice assistant — alongside a new family of five Apple Foundation Models (AFM 3) custom-built in collaboration with Google. The keynote marked Apple’s most aggressive AI push to date, turning Apple Intelligence from a set of features into a full operating system layer.

Siri AI now features a dedicated app that syncs across all Apple devices, multi-turn conversational depth, screen awareness, personal context integration across messages, email, and photos, and real-time web knowledge. It is powered by a new model architecture that routes queries intelligently: simple requests run on-device via the 3B-parameter AFM Core or the 20B-parameter AFM Core Advanced (which activates just 1-4B parameters per request using MoE), while complex queries escalate to Private Cloud Compute running on NVIDIA GPUs inside Google Cloud.

The Foundation Models Framework opens Apple’s on-device AI to third-party developers, allowing them to build generative AI features into apps with zero inference cost and full offline capability. Apple also announced that Claude support is coming to the Foundation Models framework on iOS 27, macOS 27, and visionOS 27, enabling third-party developers to call Claude through Apple’s native AI framework. The new models support up to 256K context windows and span over 140 languages.

Why it matters: By embedding AI at the OS level, subsidizing on-device inference, and opening its model framework to developers — all while maintaining its privacy architecture — Apple has fundamentally changed the competitive landscape for AI-powered applications on consumer devices.