Apple's WWDC 2026 Developer AI Overhaul: Free Foundation Models, Gemini Partnership, and Xcode 27 Agents

The Google Gemini partnership goes deeper

Apple and Google formalized their multi-year collaboration, with Apple paying approximately $1 billion annually for a custom 1.2-trillion-parameter Gemini model that now serves as the brain of the rebuilt Siri . On the developer side, the integration shows up in two places. First, Gemini models are callable directly through the Foundation Models framework using the same LanguageModel protocol as Apple’s first-party models . Second, Google released Firebase AI Logic, which lets developers add Gemini-powered features to an app without standing up a backend server .

The joint work extends into Xcode 27, where Gemini now appears as a built-in agentic coding assistant alongside Anthropic’s Claude and OpenAI’s ChatGPT . These coding agents handle complex tasks—writing tests, interactive planning, and autonomous ideation—not just single-line autocomplete . This is a signal that Apple views the IDE experience as a competitive surface for AI, not just the apps developers produce.

Xcode 27: Three agents, one IDE

Xcode 27 ships with three plug-in coding agents as first-party integrations: Gemini from Google, Claude from Anthropic, and ChatGPT from OpenAI . This is distinct from previous code-completion features. Apple is labeling these as "agentic coding" tools, meaning they can reason across files, plan multi-step refactors, and execute tasks within a new Device Hub testing environment .

The underlying model routing is connected to the same Foundation Models framework developers use in their own apps, which means Apple can add new model providers later by extending the LanguageModel protocol without rebuilding the Xcode agent interface .

The free tier that changes the economics

The pricing change for small developers is the most structurally significant shift on the App Store side. Apple confirmed that developers with fewer than 2 million total first-time downloads can access Foundation Models on Private Cloud Compute at no cost, with no API keys or account setup required . The server-side model supports a 32K context window with configurable reasoning levels .

This removes what had been the single largest obstacle for indie shops: the unpredictable cloud inference bill that scales with usage. By absorbing that cost, Apple is betting that more developers will build AI-native features, increasing the stickiness of the platform . The offer is tied to the existing App Store Small Business Program, so enrollment is a prerequisite .

On-device advances: from three lines of Swift to systemwide search

Apple continues to push its on-device model as a privacy-first workhorse. Developers can call it with as few as three lines of Swift . The model now accepts images natively, which removes the earlier limitation of running a separate Vision model as a pre-processing step . Apple’s developer documentation also highlights new built-in tools: a BarcodeReaderTool, an OCRTool backed by Vision, and a Spotlight-powered search tool that enables on-device RAG without any network round trip .

Alongside the Foundation Models expansion, Apple introduced the Core AI framework, a lower-level on-device interface optimized for Apple Silicon, and a pair of scripting tools—a command-line fm utility and a Python SDK—for building AI-powered workflows outside the IDE .

What is not yet available everywhere

The rebuilt Siri, powered by the custom Gemini model, entered developer beta with iOS 27 and macOS 27 (code-named Golden Gate) on June 8, 2026 . However, Apple confirmed that regulatory constraints mean the Gemini-powered Siri will not launch in Europe and China at launch . The framework itself is still marked as beta in developer documentation for iOS 26 and watchOS 27, so some API surfaces may change before the public release this fall .