Apple to host third‑party AI models on iPhones for on‑device Apple Intelligence

Apple is not newly “hosting third-party AI models” on iPhones. What Apple actually shipped is a way for third-party apps to use Apple’s own on-device model through a new Foundation Models framework. (apple.com) (developer.apple.com) That framework was announced at WWDC on June 9, 2025, when Apple said developers would get direct access to the large language model at the core of Apple Intelligence. Apple said the model is “powerful, fast, built with privacy, and available even when users are offline.” (apple.com) In plain terms, a foundation model is the general-purpose text engine behind features like summarizing, rewriting, or extracting information from text. Apple’s version runs on the device itself, so an app can ask the model for help without sending every request to a remote server. (developer.apple.com 1) (developer.apple.com 2) Apple says developers can plug into that model with native Swift support and, in some cases, “as few as three lines of code.” The framework supports text generation, summarization, structured output and “tool calling,” which lets the model trigger app code or query a service for extra information. (developer.apple.com 1) (developer.apple.com 2) That is different from letting developers freely swap in outside large language models as first-class system models on iPhone. The official Apple material describes access to Apple’s on-device model, not a marketplace for OpenAI, Google, Anthropic or other third-party models running natively inside Apple Intelligence. (developer.apple.com) (apple.com) Apple does allow developers to customize behavior with adapters, which are smaller add-ons trained for a specific app task or domain. Apple’s adapter toolkit teaches the system model “new skills specific to your app,” but each adapter is tied to a specific system model version and requires a separate entitlement to deploy. (developer.apple.com) Apple is also mixing on-device and cloud AI in other places. In Shortcuts, Apple says a new “Use Model” action can tap Apple Intelligence models either on device or through Private Cloud Compute, depending on the task. (developer.apple.com 1) (developer.apple.com 2) The hardware boundary matters. Apple says the Foundation Models framework requires iOS 26, iPadOS 26, macOS 26 or visionOS 26, and Apple Intelligence itself only runs on supported devices that have the necessary chips and settings enabled. (developer.apple.com) (apple.com) Apple’s pitch is straightforward: keep more AI work local, cut latency, and avoid shipping sensitive prompts to the cloud by default. The company is trying to make Apple Intelligence something app developers can embed directly, while keeping the core model under Apple’s control. (apple.com) (developer.apple.com)