// 2026.06.12

WWDC 2026 for XR Builders — visionOS 27, RealityKit Splats, and the Spatial Web

10:00 JST

Apple’s WWDC 2026 keynote landed on Monday, June 8, and the developer betas are live now — which means the visionOS 27 SDK is installable today, two and a half weeks before the hackathon. Several of this year’s headline APIs map directly onto what people build here, on Apple Vision Pro and on the web. This is a developer’s-eye filter on the announcements: what is genuinely new, what is worth your time at the end of June, and what is rumor or last year’s news wearing a fresh label.

One useful pointer first: the session most relevant to this crowd is “Explore advances in RealityKit” (WWDC26 #279) — the deep dive behind most of the spatial APIs below.

▸ visionOS 27 — what’s new · WebKit in Safari 27 beta · What’s new in Xcode 27

The lineup, briefly

Apple kept the year-aligned numbering from last cycle: everything is now 27 — iOS 27, iPadOS 27, macOS 27, visionOS 27, Xcode 27, Safari 27. The new Swift is 6.4 (not 7). Developer betas shipped the day of the keynote; the public beta is expected in mid-July, with general release in fall 2026. For a late-June hackathon that means the SDK and simulator are available now, but anything you ship publicly still targets the current shipping OS.

RealityKit renders Gaussian splats now

The single most relevant change for this series: RealityKit on visionOS 27 renders 3D Gaussian splatting natively, via a GaussianSplatResource and GaussianSplatComponent. The splat pipelines covered here all week — SHARP and LiTo turning photos into splats, Scaniverse capturing them on a phone — have until now reached Vision Pro only through a third-party Metal renderer or by flattening the splat to a mesh and losing what made it a splat. As of visionOS 27, a .ply capture drops into a RealityKit scene as a first-class entity. (Gaussian splats on visionOS, session #279.)

The same session brought a batch of rendering APIs worth knowing:

  • Physical Space Lighting — virtual lights cast onto real surfaces.
  • Projective Textures — textures projected from a spotlight (stained-glass, caustics).
  • Real-time cloth simulation — ClothBodyComponent / ClothColliderComponent, with kinematic pinning.
  • Reverb Mesh — physically modeled spatial audio from room materials.
  • Navigation Mesh — pathfinding for entities, with off-mesh links and traversal costs.
  • Lightmaps and soft shadows, mesh LOD by screen size, thermal-aware rendering.

And the framing message from Apple is cross-platform: RealityKit content now targets visionOS, iOS, iPadOS, macOS, and tvOS from one codebase, and ARKit object tracking expanded to iOS. If you build the spatial scene in RealityKit, more of it travels beyond the headset than before.

The spatial web: read this before you reach for WebXR

This is where it pays to be precise, because the obvious assumption is wrong. WebXR is not new on Vision Pro — Safari has supported immersive-vr sessions by default since visionOS 2 (2024), using gaze-and-pinch input. And the WebXR AR module (immersive-ar) is still non-functional on Vision Pro: the feature flag exists, but Apple’s engineers confirm it is not in a testable state. So for full-VR web sessions, WebXR works and is unchanged; for AR-style “3D on a web page,” WebXR is not the path.

The actual 2026 web story is the HTML <model> element, and it is a strong one for a WebXR-minded audience (WebKit in Safari 27 beta, session #204):

  • <model> renders USDZ assets at real-world scale, and it now ships on iOS, iPadOS, and macOS too — not just visionOS.
  • A new Immersive API (it behaves like the Fullscreen API) lets one JavaScript call open an immersive presentation of a model, anchored at the user’s feet, emerging from behind the Safari window. Using display:none defers the asset download until that request fires.
  • Web Environments are enabled by default in Safari 27 on visionOS 27 — a site can supply an immersive background.
  • The <img> element gains a controls attribute for viewing spatial and panorama photos.

For a team that wants 3D content reachable from a URL on any Apple device, <model> is now the robust, supported route — and it is a cleaner story than it was a year ago. WebXR remains the choice for genuinely immersive, interactive VR sessions and for the cross-headset work targeting Quest 3 and Spectacles.

Engines, accessories, and streaming

A few more developer-facing moves matter for hackathon tooling:

  • Compositor Services is now adoptable by game engines on visionOS and macOS. Unity, Unreal, and Godot can use it as a rendering path; Godot gains a RealityKit rendering plugin and a PHASE audio plugin. That is directly relevant to the Godot and Unity PolySpatial routes covered here.
  • The Spatial Accessories framework opened to third-party hardware — IR-LED + IMU tracked controllers, 6DoF at up to 90 Hz, with buttons, touchpad, and haptics. Unity adds PSVR2 Sense controller support through it.
  • Foveated Streaming (available from visionOS 26.4) and the new Spatial Preview framework let a Mac or workstation stream heavy immersive content — including live USD scenes — to the headset without a separate visionOS app. Apple demoed it driving X-Plane 12 with ARKit aligning the virtual cockpit to physical hardware.
  • Apple Immersive Video gains a universal .aivu file format (90 fps, 100+ megapixels per eye, Apple Spatial Audio), and Reality Composer Pro 3 adds live on-device preview.

Xcode 27 and the agent story

The agentic-coding piece earlier this week noted that the interesting AI-tooling trend is provider-swappable agents, and Apple moved the same direction. Xcode 27 keeps the Claude and OpenAI coding integrations from Xcode 26 and adds Google Gemini, plus on-device predictive completion that runs on the Apple Silicon Neural Engine, and an agentic workflow built around editable “plan” artifacts you approve before the agent acts. Separately, Apple’s Foundation Models framework is reported to add a model protocol that lets an app target Apple’s on-device model or third-party models like Claude and Gemini through a common interface, now with image input. The provider-swap details are still thinly sourced outside Apple’s own sessions — confirm against session #258 before relying on specifics — but the direction is unmistakable.

One caveat for an international audience: the rebuilt AI Siri does not ship in the EU or China with this release, but the Foundation Models developer API is not geo-restricted.

What did not happen — and what is just rumor

  • No new Vision Pro hardware, and no smart glasses. The M5 Vision Pro refresh already shipped in October 2025; nothing new was announced at WWDC. Apple’s rumored display-less glasses (codename “N50”) are reported to have slipped to late 2027, with a higher-end model later — all rumor, none of it announced. The read worth taking away is that visionOS 27’s “Visual Intelligence” and spatial-Siri work is the software groundwork for whatever glasses eventually arrive.
  • Mind the carryover. Some recaps credit visionOS 27 with the volumetric Persona overhaul and the enterprise main-camera / Insta360 APIs — both of those were visionOS 26 (2025), not new this year. Don’t plan a project around them as “new.”
  • Simulator limits. The Xcode 27 beta includes the visionOS 27 simulator, but Gaussian splats, the new lighting APIs, object tracking, and spatial accessories need a physical Vision Pro to test. Budget for device time if your hack leans on them.

Questions? Reach the team via the Contact page.

// BACK TO NEWS