Skip to content

Black Friday: 40% off. Purgy for $12 $20 with code Code , until Dec 12.

Get the deal
← Blog

How to Delete Ollama Models on Mac and Free the Space

On this page
  1. How to delete an Ollama model, step by step
  2. What really happens to the files when you delete a model
  3. Delete every model, or uninstall Ollama
  4. Delete a model through the API
  5. Questions about deleting Ollama models
  6. See every AI model on your Mac in one place

Quick answer: Run ollama list to see your models, then ollama rm followed by the model's name. Ollama deletes that model and any files only it uses. The Ollama server must be running first, so open the app or start it from Terminal.

Tested on: MacBook Air M1, 16 GB, macOS 27.0.1, Ollama 0.34.2 installed with Homebrew, on 19 November 2026. I ran every command in this guide. To protect my real models, I tested deletion on a copy and on a throwaway custom model.

How to delete an Ollama model, step by step

These steps work for the Ollama app and for a Homebrew install. They all happen in Terminal.

  1. Make sure Ollama is running. The commands talk to the Ollama server. With the server off, ollama --version on my Mac printed "could not connect to a running Ollama instance". Open the Ollama app, or run ollama serve in a second Terminal window. A Homebrew install can also use brew services start ollama.
  2. List your models. Run ollama list. This is what it printed on my Mac:
    NAME           ID              SIZE      MODIFIED
    llama3.2:3b    a80c4f17acd5    2.0 GB    8 weeks ago
    qwen3.5:4b     2a654d98e6fb    3.4 GB    8 weeks ago
  3. Remove the model. Run ollama rm and copy the name exactly as shown in the first column, tag included, for example ollama rm llama3.2:3b. Ollama answers with a line like deleted 'purgy-test-copy', which is the throwaway model I removed in my test.
  4. Check the space. Run du -sh ~/.ollama/models before and after to see what you got back.

Ollama's documentation also lists ollama ps to show which models are loaded, and ollama stop to unload one. By default a model stays in memory for 5 minutes after you use it.

What really happens to the files when you delete a model

Ollama stores each model as small manifest files plus large data files called blobs. Models can share blobs. So the real question is whether deleting one model can hurt another, and whether it frees any space.

I tested both on my Mac with two harmless models and counted the files in ~/.ollama/models/blobs at each step:

StepFiles in blobsBoth real models still listed?
Start10Yes
Copy a model with ollama cp10 (no new data)Yes
Remove that copy with ollama rm10Yes
Create a custom model with its own system prompt12 (two tiny new files)Yes
Remove the custom model with ollama rm10 (the two new files gone)Yes

Three things follow from this:

  • A copy costs almost nothing. ollama cp added one small manifest and no data.
  • Removing a copy is safe. The files it shared with the original stayed where they were.
  • Ollama removes the files only that model uses. The custom model's two new files were deleted, and everything else was left alone.

I did not delete either of my real models, so I cannot show you the exact gigabytes returned. Measure your own folder before and after. It is the only number that matters on your disk.

Delete every model, or uninstall Ollama

All models live in ~/.ollama/models, so deleting the whole ~/.ollama folder removes every model at once. The folder also holds Ollama's key files. On my Mac they are id_ed25519 and id_ed25519.pub.

For the Ollama app, Ollama's documentation says to remove these to uninstall it fully:

sudo rm -rf /Applications/Ollama.app
sudo rm /usr/local/bin/ollama
rm -rf "$HOME/Library/Application Support/Ollama"
rm -rf "$HOME/Library/Saved Application State/com.electron.ollama.savedState"
rm -rf "$HOME/Library/Caches/com.electron.ollama"
rm -rf "$HOME/Library/Caches/ollama"
rm -rf "$HOME/Library/WebKit/com.electron.ollama"
rm -rf "$HOME/.ollama"

These are the paths in Ollama's macOS documentation. I wrote them with $HOME, because a tilde inside quotes is not expanded by Terminal. Check each path before you press Return, because rm -rf has no undo and no Trash.

That documentation does not cover Homebrew. On my Mac, Ollama is a Homebrew formula: the command is at /opt/homebrew/bin/ollama and there is no Ollama.app. For that setup the normal Homebrew route is brew uninstall ollama, then removing ~/.ollama. I did not run it, because I still use Ollama.

Delete a model through the API

If you manage models from a script, Ollama's API has a delete call. It is an HTTP DELETE request to the local server, with the model name in a JSON body:

curl -X DELETE http://localhost:11434/api/delete -d '{ "model": "llama3.2:3b" }'

The API reference lists a 200 response when the model is deleted.

Questions about deleting Ollama models

These are the questions that come up most.

Can I delete the files in Finder instead?

I would not. In my folder the model data is ten files named by checksum, such as sha256-81fb60c7…. Nothing in a name says which model owns it, and a manifest could still point at a file you removed. I did not test what Ollama does then. Use ollama rm.

Will I lose anything I made?

Models you downloaded can be downloaded again with ollama pull. A custom model's system prompt and settings exist only on your Mac. Save them first with ollama show --modelfile your-model > your-model.Modelfile.

Why does ollama list say 5.4 GB when du says 5.0G?

They count differently. My two models total about 5.4 billion bytes. Ollama reports that as 5.4 GB. du -sh uses units of 1,024, so the same bytes show as 5.0G. Nothing is missing.

Where is the biggest model data on my Mac?

Look at the SIZE column of ollama list first. Then check other tools. On my Mac the Hugging Face cache is 11 GB, more than twice the size of all my Ollama models together.

See every AI model on your Mac in one place

Ollama is rarely the only place models end up. LM Studio, Hugging Face's cache and GPT4All each keep their own files, and they do not share them with Ollama.

Purgy's AI models screen lists them together, each tool with its total size. Nothing is ticked for you, because re-downloading a large model costs time. Anything you choose goes to the Trash, so Finder can put it back.

For one Ollama model, ollama rm is still the most precise tool. Purgy is for the wider picture. For where the files live, see where Ollama stores models on Mac. For planning disk space, see how much storage local LLMs need. You can scan your Mac with the free trial.