Where does Ollama store models on Mac?

In ~/.ollama/models. Run ollama list to see them and ollama rm to delete one; Ollama only removes the pieces no other model uses.

Can I delete the Hugging Face cache?

Yes, every model in ~/.cache/huggingface can be downloaded again. huggingface-cli delete-cache lets you choose which ones to delete instead of removing everything.

How much space does a local AI model need?

Roughly 4–5 GB for a 7–8B model, about 20 GB for a 30B model and 40 GB or more for a 70B model, depending on the quantization.

AI models filling your Mac? Where Ollama, LM Studio and Hugging Face keep them

In short

Every local AI model is a multi-gigabyte download, and they pile up in hidden folders: ~/.ollama, ~/.lmstudio and ~/.cache/huggingface. They can all be downloaded again, so the only question is which ones you still use. Delete them with each tool's own command, not by hand.

Why models take so much space

A model is mostly its weights: billions of numbers. A small 7–8B model takes about 4–5 GB, a 30B one around 20 GB and a 70B one 40 GB or more. Trying a few models to compare them is enough to fill 100 GB, and nothing deletes the ones you stopped using.

They don't show up in Finder because they live in hidden folders in your home folder, so in Storage settings they usually count as “System Data”.

Ollama

Models are in ~/.ollama/models. To see them with their size and when they were last changed:

ollama list

And to delete one:

ollama rm <model>

Use ollama rm rather than deleting files: models share pieces (layers), and Ollama only removes the ones no other model uses. If you'd rather keep them on an external drive, recent versions of the Ollama app let you change the model location in Settings; otherwise, set the OLLAMA_MODELS environment variable.

LM Studio

Models are in ~/.lmstudio/models (older versions used ~/.cache/lm-studio/models). The easiest way to clean them is from the app: open the My Models tab, which lists each model with its size, and delete the ones you don't use from its menu. You can also move the models folder from there.

Hugging Face

Anything that downloads models with the Hugging Face libraries (transformers, diffusers, many Python scripts and notebooks) keeps them in ~/.cache/huggingface/hub. To see what's there, with sizes and when each one was last used:

huggingface-cli scan-cache

And to pick which ones to delete from a list:

huggingface-cli delete-cache

In recent versions the command is called hf; hf --help lists its cache commands. To keep the cache somewhere else, set HF_HOME.

Other places models hide

Whisper (speech to text): ~/.cache/whisper.
PyTorch: ~/.cache/torch, weights downloaded by code that uses it.
Chroma: ~/.cache/chroma, embedding models.
Chrome downloads Gemini Nano by itself to ~/Library/Application Support/Google/Chrome/OptGuideOnDeviceModel. If you delete it, Chrome downloads it again unless you turn off the optimization-guide-on-device-model flag in chrome://flags.
Image apps like ComfyUI keep their models in their own models folder; Draw Things keeps them inside its folder in ~/Library/Containers.

Apple Intelligence also needs about 7 GB, but macOS manages those models: you can't delete them by hand.

Before you delete

Everything can be downloaded again, but it takes time and bandwidth: think of a 40 GB model on a slow connection.
Quit the app first (Ollama, LM Studio, the notebook) so it isn't using the model.
Fine-tuned models or adapters you made yourself are not downloads: if you trained them, back them up before cleaning.

DiskSpace AI does this for you

It analyzes your Mac in minutes, tells you what each folder is and only selects what recreates itself. Your files are kept 7 days in case you change your mind. Free during the beta.

Download the Beta

Folders in this guide

Ollama models~/.ollama/modelsReview
LM Studio models~/.lmstudio/modelsReview
Hugging Face models~/.cache/huggingfaceReview
Whisper models~/.cache/whisperReview
PyTorch models~/.cache/torchReview
Chroma models~/.cache/chromaReview
Chrome AI model~/Library/Application Support/Google/Chrome/OptGuideOnDeviceModelReview
Claude virtual machine~/Library/Application Support/Claude/vm_bundlesReview