WWDC 2026: Apple Intelligence 2.0 and the New Developer APIs

Apple's WWDC 2026 was framed around a single organizing theme: the transition of AI from discrete features to ambient infrastructure woven through every part of the platform. Apple Intelligence 2.0 — the updated suite of on-device and cloud-backed AI capabilities — anchors the OS releases across iOS 20, macOS 16, and visionOS 4. For developers, the conference was dense with new APIs, and the ratio of immediately useful to aspirational-but-experimental skewed more practical than in some previous years.

Apple Intelligence 2.0: What Changed

The original Apple Intelligence, released with iOS 18, was primarily a writing, summarization, and Siri improvement story. The model quality was solid but the capability scope was limited — there was no multimodal input to speak of, and the integration with third-party apps was constrained to predefined action categories.

Apple Intelligence 2.0 extends the on-device model to multimodal inputs natively. The Apple Neural Engine in M4-series chips (iPhone 17, M4 Macs) runs a substantially larger model than the original — Apple hasn't published parameter counts but has described it as "multiple times larger" than the 3B-parameter model used for on-device inference in Apple Intelligence 1.0. The immediate practical effect is more reliable text understanding, better context retention in extended Siri conversations, and the ability to understand images natively without the previous image-description intermediate step.

Private Cloud Compute — Apple's privacy-preserving server inference infrastructure, where queries to larger models are processed on Apple Silicon servers with cryptographic guarantees that Apple cannot see the content — has been expanded. Apple Intelligence 2.0 routes more complex requests to Private Cloud Compute while maintaining the principle that requests are never logged or used for training.

The capabilities developers care about most: longer context handling (practical 32K context window in the on-device model for most document-length inputs), better code understanding (the Writing Tools API now supports code-aware formatting and refactoring suggestions), and improved intent recognition for App Intents.

Foundation Models Framework

The most significant developer API announcement is the Foundation Models framework — a direct API to the on-device Apple Intelligence model, available to third-party apps for the first time.

Previous Apple Intelligence developer access was mediated through higher-level APIs: Writing Tools (text transformation), App Intents (action execution), and Image Playground (image generation). The Foundation Models framework exposes the underlying model directly, allowing developers to:

• Make text completion and instruction-following calls against the on-device model with custom system prompts
• Perform structured output generation (JSON mode) with schema validation
• Run multimodal queries combining text and images

The framework is gated behind user permission and requires a privacy disclosure in the app's entry in the App Store. The model runs entirely on-device; data never leaves the device through the Foundation Models API (as opposed to requests that opt into Private Cloud Compute, which is a separate path). This has obvious privacy appeal for categories like health, finance, and notes apps.

The limitation worth noting: the Foundation Models API doesn't expose the full model used for Apple's own features. It's a smaller, more efficient variant. Apple hasn't published exact specifications, but developer early access testing suggests context length constraints around 8K tokens for the Foundation Models API, compared to 32K for Apple's own features. This may change as Apple Silicon improves.

Swift 6.2

Swift 6.2 reduces the friction of the strict concurrency model introduced in Swift 6.0. The original Swift 6 concurrency changes were correct in principle — structured concurrency with complete actor isolation and Sendable checking eliminates data races at compile time — but they generated significant migration pain for existing codebases. Swift 6.2 introduces module-level concurrency defaults that allow gradual opt-in to strict checking rather than requiring immediate full compliance.

The language update also includes typed throws with associated error types (allowing throws(NetworkError) instead of untyped throws), which makes error handling in Swift more expressive and reduces the need for catch clauses that re-cast errors.

For Foundation Models API integration specifically, Swift 6.2 ships with concurrency-safe bindings that eliminate the boilerplate of wrapping async Foundation Models calls in actors.

Xcode AI Features

Xcode 18 ships with two significant AI additions:

Swift Assist (the renamed and significantly improved Xcode AI code completion) now uses the Foundation Models API and a cloud component to provide context-aware suggestions that understand your project structure, not just the open file. It's competitive with GitHub Copilot in code completion quality for Swift and Objective-C, and has the advantage of deep Xcode integration — it understands build errors, connects completions to documentation, and can explain warnings inline.

AI-powered test generation is a new feature in Xcode 18. Select a function, invoke Generate Tests, and Xcode produces a test case with representative inputs and expected outputs. The quality varies — it's better for pure functions than for complex UI or networking code — but it's a useful starting point and reflects the same pattern we're seeing in other IDEs.

visionOS 4

For the small but growing visionOS developer community, visionOS 4 brings better window management APIs, significantly improved hand tracking accuracy for custom gesture recognition, and the Spatial AI framework — a new API that makes it easier to integrate on-device vision models that understand the physical environment.

The platform story for visionOS is still being written. Apple Vision Pro adoption is real but limited by price. The developer investment in visionOS apps is proportionally small compared to iOS. visionOS 4's improvements are meaningful for existing developers but unlikely to dramatically shift investment until the device price comes down.

What Developers Can Actually Build Now

Concretely, the APIs announced at WWDC 2026 enable:

• Notes and journaling apps that summarize, categorize, and query entries using the Foundation Models API without any cloud dependency
• Health and wellness apps that analyze user data and provide personalized recommendations fully on-device
• Productivity apps with context-aware automation via improved App Intents
• Developer tools built in Swift that use AI-powered refactoring

The through-line is on-device inference as a first-class platform capability. Apple's privacy differentiation from cloud-first AI products is becoming a concrete API story, not just a marketing position.