When to use Core ML

Core ML takes a trained model and runs it in an iOS or macOS app. Apple still lists it for traditional machine learning. That includes tree models, regression, and other classic jobs. Apple has not announced a deprecation.

You convert the model with Core ML Tools, often called coremltools. The usual file is a .mlmodel. You add that file to Xcode. The app loads it through the Core ML framework.

import coremltools as ct

model = ct.convert(trained_model)
model.save("classifier.mlmodel")

That snippet is the public convert-and-save path. Check the Core ML Tools docs for the model type you have. Do not copy an argument you have not checked.

What the OS does

After the file is in the app, the system picks much of the run path. Core ML can use the CPU, the GPU, or the Neural Engine. You do not write a full custom runtime. That is useful when you want a stable shipping path and you do not need a new generative stack.

When Core ML is the right pick

• The app must run on an OS older than 27.
• The model is a classifier, a detector, a tree model, or another classic ML job.
• You already have a Core ML file and it does the job.

Apple's machine learning page still points classic ML at Core ML. For large language models and other generative models, that same page points you to Core AI. Keep Core ML when it matches the app. Apple still supports Core ML.

Model architectures in plain English shows the classifier shape. PyTorch to Core AI in Xcode is the Core AI course for a custom network on 27 or later. Core AI models, typed lists ready-made Core AI presets, not Core ML files.

Next, look at Core AI, the current path for a custom neural model on 27 or later.