AFM 3 Core Powers Apple's Native Bill-Splitting Camera Tool
Apple is adding a 20-billion parameter multimodal model to iOS 27, allowing the native Camera app to scan receipts and process Apple Cash split payments.
During the June 8 WWDC 2026 keynote, Apple introduced a native bill-splitting tool powered by a new Siri mode in the Camera app. Part of the broader Visual Intelligence suite arriving in iOS 27, the feature interprets physical receipts in real-time to itemize dining bills and route payment requests directly through the operating system.
On-Device Visual Processing
The camera integration relies on the AFM 3 Core Advanced model. This is a 20-billion parameter multimodal model built with a sparse architecture, designed specifically for on-device efficiency. During inference, the system activates only 1 to 4 billion parameters per request. This approach preserves battery life and minimizes latency when mapping physical line items on a receipt to digital text elements on the screen. By controlling how model parameters are engaged natively, Apple can process text and numerical extraction without pushing image data to external servers.
Apple Vice President of Software Engineering Sebastien Marineau-Mes framed the update as a mechanism to take immediate action on physical environments. In practice, the camera identifies individual items on a restaurant bill and renders them as selectable interface elements. Users tap the items they ordered, and the system automatically calculates proportional tax and tip allocations.
Financial and Hardware Integration
The technical advantage of Apple’s implementation lies in its end-to-end payment routing. Once the bill is divided, the camera mode hands the numerical data directly to Apple Cash. The system generates exact payment requests and distributes them to contacts via Messages or Wallet. This eliminates the need for manual data entry or third-party splitting applications, embedding the entire lifecycle from optical character recognition to peer-to-peer transaction inside the native camera view.
The feature requires an iPhone 11 or newer. Apple is also shipping the capability to visionOS 27, where it manifests as a floating Siri orb that analyzes objects in the wearer’s physical environment. All visual analysis is governed by Private Cloud Compute, ensuring that sensitive receipt data and personal transactions remain encrypted. This localized execution mirrors the privacy constraints seen as Apple shifts Siri toward a broader agentic architecture.
Release Constraints
Developer betas containing the new camera mode are available immediately, with a public release slated for Fall 2026. The rollout is subject to strict regional and linguistic constraints. The feature will launch exclusively in English. Furthermore, Apple indicated that regulatory friction regarding the Digital Markets Act will delay or entirely block the deployment of these AI features for users in the European Union.
For developers building financial utilities or shared-expense applications, this update represents a significant shift in the iOS ecosystem. If your application relies on receipt scanning or basic tab calculations, the operating system now handles this friction natively. You should evaluate how your app provides value beyond point-of-sale math and focus on complex expense tracking or cross-platform synchronization that Apple Foundation Models do not currently address.
Get Insanely Good at AI
The book for developers who want to understand how AI actually works. LLMs, prompt engineering, RAG, AI agents, and production systems.
Keep Reading
How to Call Claude 4.5 via Apple Foundation Models Framework
Learn how to integrate Claude 4.5 into your Swift applications using Apple's new Foundation Models framework for hybrid on-device and cloud processing.
Native iOS 27 Workloads Can Now Route to Claude and Gemini
Apple's Extensions framework for iOS 27 allows developers to integrate third-party AI models directly into native Siri and Writing Tools workflows.
Google Drops Vision Encoders in Gemma 4 12B Multimodal Release
Google DeepMind's new 12-billion parameter model uses a unified architecture to process text, image, and native audio directly on laptops with 16GB of RAM.
How to Run Gemma 4 On-Device with LiteRT-LM
Learn how to configure LiteRT-LM to deploy the Gemma 4 model family locally across mobile, desktop, and edge environments with constrained JSON decoding.
AI Edge Gallery for Android Gains On-Device MCP and Gemma 4
Google updated the AI Edge Gallery Android app with experimental Model Context Protocol support, enabling on-device Gemma 4 models to use external web tools.