Buy Pro

Guides AI models

Where local AI models are stored on a Mac, and how to remove them

Ollama, LM Studio and Hugging Face each keep models in a different hidden folder, and one model can be 40 GB. Here's where to find them and how to remove them properly.

Updated 17 November 2026

Why they take so much space

Running AI on your own Mac means downloading the whole model. A small model is about 4 GB, and a large one can be 40 GB or more. Each app keeps its own copy in its own hidden folder, so trying a few is enough to fill a disk.

To see how much they use, open Terminal and run:

du -sh ~/.ollama ~/.lmstudio ~/.cache/lm-studio ~/.cache/huggingface ~/.cache/torch ~/.cache/whisper 2>/dev/null

Folders you don’t have are skipped.

Where each app keeps its models

App or library Folder
Ollama ~/.ollama/models
LM Studio ~/.lmstudio/models (older versions: ~/.cache/lm-studio/models)
Hugging Face ~/.cache/huggingface/hub
PyTorch Hub ~/.cache/torch
Whisper ~/.cache/whisper

Some of these can be moved. Ollama uses the folder in OLLAMA_MODELS if it’s set, and Hugging Face uses HF_HOME.

Ollama

Don’t delete files in ~/.ollama/models by hand. Ollama stores models in shared pieces, so one file can belong to several models. Use Ollama’s own commands instead. Ollama needs to be running.

List your models and their sizes:

ollama list

Remove one:

ollama rm llama3.1:70b

This deletes the model straight away. It doesn’t go to the Trash. To get it back, download it again with ollama pull.

LM Studio

Open LM Studio, go to My Models, and delete the ones you don’t use. You can also quit LM Studio and move a model’s folder to the Trash. Models are grouped by who published them, such as lmstudio-community/.

Hugging Face

Python libraries such as transformers save each model they download to ~/.cache/huggingface/hub. Each one has its own folder, named like models--meta-llama--Llama-3.1-8B. Datasets are in folders starting with datasets--.

Move the folders for models you no longer use to the Trash. If your code needs one again, it downloads it again, which can take a while for a large model.

PyTorch and Whisper

~/.cache/torch and ~/.cache/whisper hold models downloaded by PyTorch Hub and OpenAI’s Whisper. They’re safe to move to the Trash. Code that needs them downloads them again.

AI apps’ caches

Desktop AI apps such as Claude, Cursor and Windsurf keep web and graphics caches in ~/Library/Application Support, inside each app’s folder. Folders named Cache, Code Cache and GPUCache are safe to clear after you quit the app.

Leave the rest of those folders alone. They can hold your conversations, settings and sign-in.