Apple Developer Weekly #248 Running AI Models on iPhone
Final reminder: 13+ annual plan offer available until today
From 13
From the A19 Pro hardware performance, Foundation Models official launch, Swift Transformers 1.0 release, and more, I truly feel that the era of running AI models on iPhone has arrived. At least the difficulty of developing local AI applications is rapidly decreasing.
What do you think?
13 on September 30, 2025
➕13+ Recent Content Updates
Final reminder: 13+ annual plan offer available until today.
Offering an annual plan represents my commitment to continue writing for another year, which is a very serious consideration. If you want to subscribe, don’t miss out.
👉Start Your Learning Journey for the Next Year (click here to open Patreon).
Last week’s ⚡️Exploring iPhone USB Speed Impact on Development Experience series, the fifth article “AI Models Need High-Speed Transfer“ is the most important.
Starting with the iPhone 17 generation, running AI models on device has shifted from feasible to practical. Developing AI applications requires frequent transfer of multi-GB models, and speed bottlenecks will be a brand new challenge for workflows and development efficiency.
Plus monthly review and a development habits reminder:
💡Good Development Habits: Reading Release Notes From a small pitfall in Apple documentation, emphasizing the importance of reading Release Notes for avoiding issues
🤖Code Along with Foundation Models
Foundation Models changed a lot during beta, making the WWDC25 video outdated.
This week Apple released the latest Code Along video.
Bonus: Apple press release introducing many real-world app cases using Foundation Models.
Traditional Chinese Apple Intelligence is available in iOS 26.1 beta 1, making now a great time to get started with Foundation Models.
🚀Swift Transformers 1.0
The very night I published the article AI Models Need High-Speed Transfer, Hugging Face announced the official release of Swift Transformers 1.0 and outlined future development.
This package aims to reduce the hassle for developers using local AI models on Apple platforms, filling gaps in Core ML and MLX. It can be said to be an essential tool for running third-party open-source AI models on Apple platforms.
In my view, Swift Transformers and Foundation Models share similar essence—both make it easier for developers to use AI models in Apple platform apps.
It has three main components:
Tokenizers: Tools for preparing input parameters for models
Hub: Tools for downloading and managing models
Models and Generation: Wrappers for LLMs converted to Core ML format
Version 1.0 also includes many key updates, including more independent modular design, Swift Jinja integration, and full Swift 6 support.
🛠️Swift Configuration
Many developers should be happy—Swift officially introduces a new solution for managing settings and environments. I think this is something with limitless potential.
There’s one “small downside”—iOS requires 18 or above.
🧩SE-0494 Add isIdentical(to:) Methods for Quick Comparisons to Concrete Types
New proposal for isIdentical(to:) methods, enabling O(1) comparisons for container types like Array/String/Dictionary.
kemchenj’s tweet provides a simple explanation.
↖️Last Issue Highlights
If you’ve been too busy to read last week’s content, here are the highlights:
That’s all for this week’s Apple Developer Weekly. Feel free to like❤️, comment💬, or reply✉️ to share your thoughts.
The newsletter has always been free to read, with just occasional sponsors. Please subscribe to 13+ Premium Column to support my writing, thank you.
For business collaborations or newsletter sponsorship, please refer to this page.