How to deploy a multi-modal SLM natively on an iPhone 18

It's worth naming the problem with a headline like "deploy a multimodal SLM natively on an iPhone 18" directly: the iPhone 18 doesn't exist yet. Apple's Pro and Fold models are expected around September 2026, with standard models following in spring 2027 — so any device-specific claim about it right now would be a guess dressed up as a fact. What's real, current, and documented is Apple's Foundation Models framework, which is the actual mechanism for running a multimodal small model natively on-device on an iPhone, and it works today on current-generation hardware.

What actually shipped: multimodal input in the Foundation Models framework

The framework gives developers a native Swift API to the same on-device model that powers Apple Intelligence. Its image-input support lets you attach a photo, screenshot, or scanned document to a prompt — the model reasons about the image and text together, fully offline, with no network call and no per-request API cost. Supported image types include UIImage, NSImage, CGImage, Core Image types, and file URLs, and there's no fixed crop or aspect ratio requirement (though larger images cost more tokens and add latency).

Setup

1. Check eligibility. The framework requires an Apple Intelligence-eligible device — iPhone 15 Pro or later, with Apple Intelligence enabled in Settings — running a current iOS release with Foundation Models support.

2. Create a session and attach an image using the unified LanguageModelSession API alongside your text prompt.

3. Use the built-in vision tools where they fit — OCRTool for extracting text from an image and BarcodeReaderTool for reading barcodes — rather than asking the general model to do OCR from scratch.

// Passing an image alongside text to Apple's on-device model
// via the Foundation Models framework's unified API
import FoundationModels

let session = LanguageModelSession()

let response = try await session.respond {
    "What is written on this receipt, and what is the total?"
    Attachment(receiptImage)  // UIImage, NSImage, CGImage, or file URL
}

print(response.content)

The practical lesson here isn't about a specific future phone — it's that Apple's on-device multimodal capability is now a documented, stable API surface you can build against, with real hardware tiers to design around. When the iPhone 18 does ship, whatever Foundation Models framework version is current at that point is what will matter, not the device name itself.