Ollama Models on Mac: How to Delete Them Safely
High riskAI & machine learningUpdated
- Location
~/.ollama/models
- Risk level
- High riskLarge stores — whole dependency repositories or downloaded AI models — that take a long time to get back. Listed, never pre-selected.
- Selected by default
- No — you choose whether to include it
- What Freshist does
- Moves the selected items to the Trash, so you can undo. Nothing is deleted outright.
- In the app
- Clean → Ollama Models — “Local AI models — LARGE; Ollama re-downloads a model when you next run it.”
From Freshist's own cleaning rules — the same catalog the app uses.
What's in ~/.ollama/models
Ollama's FAQ gives ~/.ollama/models as the model location on macOS. The folder has two parts:
- blobs/: the actual model layers (weights, templates, parameters), named by their SHA-256 digest.
- manifests/: small files that say which blobs make up each model and tag, such as
llama3.2:3b.
Next to models, the ~/.ollama folder also holds an id_ed25519 key pair that Ollama uses to identify your machine to ollama.com. That isn't a model file, and you shouldn't remove it.
Why it grows
Every ollama pull, and every ollama run of a model you don't have yet, downloads it in full. Quantized 7 to 8 billion parameter models are typically 4 to 5 GB, and larger ones run to tens of gigabytes. Trying a few model families or several quantizations of the same one adds up quickly, and nothing is removed automatically.
What deleting a model costs
| Model type | After removal |
|---|---|
| Pulled from the Ollama library | ollama pull <name> or ollama run <name> downloads it again |
Created with ollama create from a Modelfile using a library model | Can be recreated if you still have the Modelfile; the base downloads again |
| Created from local GGUF or Safetensors files | Only recoverable if you still have those source files |
| Running model | Unload it first; ollama ps shows what's loaded |
Ollama's settings and the app itself aren't affected.
List installed models
ollama ls # installed models and their sizes
ollama ps # models loaded in memory right now
du -sh ~/.ollama/models
If you've set OLLAMA_MODELS, the files live in that folder instead. ollama show <model> prints a model's details, such as parameter count and quantization, which helps when you're deciding between two similar downloads.
The sizes ollama ls reports are per model. Because tags of one family can share layers, adding them up can overstate what's on disk; du on the folder gives the real total.
Remove models with Ollama
The Ollama CLI has a dedicated command:
ollama rm <model>
For example, ollama rm llama3.2:3b. It deletes the manifest and every blob no other model uses, so it's safer than deleting files from blobs by hand, which can break models that share layers. Repeat for each model you no longer need, then check ollama ls again.
Deleting the entire ~/.ollama/models folder with Ollama quit also works, but it's all or nothing.
When to keep them
A model you use every day, or one that would take hours to pull again on your connection, is worth its space. Experiments and quantizations you've since replaced are not. If you also load models from Python, the Hugging Face cache guide covers that separate store.
Sources
Frequently asked questions
Why didn't ollama rm free as much space as the model size?
Models that share layers, such as different tags of the same family, store those layers once in the blobs folder. A shared layer is only removed when no remaining model uses it.
Can I keep Ollama models on an external drive?
Yes. Set the OLLAMA_MODELS environment variable to the new folder. On macOS, the Ollama FAQ says to set it with launchctl setenv and then restart the Ollama app.
Does deleting models remove the Ollama app?
No. The app stays installed, along with the SSH key pair in ~/.ollama that identifies your machine to ollama.com. Only the model files go.