← All courses ← Course home

Lesson 01 of 05

Why people mix up the three names

Core AI, Core ML, and MLX all work with models on Apple chips. They do different work. An Edge FDE has to pick the one that matches the app.

Agent brief (llms.md)

People mix these names because they sound close and they all show up in Apple AI talk. The names are not interchangeable.

NameCore ML
NameCore AI
NameMLX

Each name is a different product. Core ML ships a classic machine-learning model in an app. Core AI ships a custom neural or generative model on iOS 27 or later. MLX trains and tests models on a Mac.

Core MLShip a classic model
Core AIShip a custom neural model
MLXTrain and test on a Mac

What happens if you pick the wrong stack

An Edge FDE ships a model on the user's device. The wrong stack can block the build. If the app must run on an OS older than 27, Core AI will not load. If you only need a classifier or a tree model, Core ML is the usual tool. If you are still training or shrinking a model on a Mac, start with MLX, then export before you ship.

AppOS, model, and output
StackCore ML, Core AI, or MLX
ResultA model that can ship

Apple also has Foundation Models. That is a Swift API for Apple's own system language model. It is not the path for bringing your own model file. This course is about the three names people mix when they bring their own model.

Read Model architectures in plain English if you want the input and output shape first. Read Core AI models, typed for Apple's ready-made Core AI presets. Read PyTorch to Core AI in Xcode when you are ready to train a small classifier and run it on the device.

Rule. Use the name that matches the work. Do not call Mac training Core AI. Do not call a shipped .aimodel Core ML.

Next, look at Core ML, the older path for putting a model in an app.

Key concepts

  • People mix the names because they sound close and all show up in Apple AI talk.
  • Core ML ships a classic machine-learning model in an app.
  • Core AI ships a custom neural or generative model on iOS 27 or later.
  • MLX trains and tests models on a Mac.
  • Foundation Models is Apple's system language model, not your weights.

Takeaways

  • You can give one sentence per tool without swapping names.
  • You can pick a stack from OS, model type, and output.
  • You will not call Mac training Core AI or a shipped .aimodel Core ML.