Ollama models taking up space on your Mac? How to delete them safely

Yes, but read this firstDiskmack safety tier: Caution

~/.ollama/models

Yes, you can delete Ollama models, but do it with the official command, ollama rm, rather than dragging folders to the Trash. Each downloaded model is a full set of LLM weights, typically 2 to 40 GB, and Ollama never removes one on its own. Deleting costs you nothing permanent: if you want a model back later, ollama pull downloads it again.

What it is

~/.ollama/models is where Ollama keeps every model you have ever pulled: Llama, Mistral, Gemma, Qwen, whatever you tried once and forgot. These are not caches or temp files. They are the actual model weights, the multi-gigabyte files that let a language model run locally on your Mac. Small models sit around 2 GB; big ones reach 40 GB or more. Inside, the folder has two parts: blobs, which holds the weight data as content-addressed layers, and manifests, which map model names like llama3:8b to the blobs they use.

It grows fast because trying models is the whole point of Ollama. Every ollama pull, and every ollama run of a model you do not have yet, adds another multi-gigabyte download, and nothing ever leaves. Pull three or four models to compare them, keep a couple of size variants around, and tens of gigabytes disappear without you noticing. On machines that run local LLMs, this is routinely one of the largest folders on the disk, and because it lives inside the hidden ~/.ollama directory, most people find it through a disk scan rather than by browsing Finder.

Is it safe to delete?

Yes, with one rule: delete models through Ollama, not through Finder. Running ollama rm llama3 removes that model's manifest and any weight layers no other model still uses. It is instant and precise. What you should not do is reach into the blobs folder and delete the biggest files by hand. Layers are shared: two variants of the same model often point at the same base blob, so removing one file can break several models at once and leave the store in a confused state.

Deleting a model loses nothing except the download itself. There is no personal data in the weights and no chat history. The one exception is models you built yourself with a Modelfile; those live in the same store, so keep the Modelfile and rebuild them if needed. The real cost of deleting the wrong thing is bandwidth and time: pulling a 20 GB model again on slow Wi-Fi is the entire punishment. Diskmack identifies this folder automatically and cleans it the safe way.

How to check its size

In Finder: In Finder, choose Go > Go to Folder (Shift-Command-G), paste ~/.ollama/models, and press Return. The .ollama folder is hidden, but Go to Folder opens it anyway. Select the models folder and press Command-I to see its total size.

In Terminal:

du -sh ~/.ollama/models

The ~ expands to your home folder automatically. For a per-model breakdown, run ollama list, which shows every installed model with its size.

How to clean it

Open Terminal (Applications > Utilities) and run: ollama list. You get every installed model with its tag and size.
Remove the ones you no longer use, one at a time: ollama rm mistral:7b, copying the exact name and tag straight from the list.
Run ollama list again to confirm, then re-check the folder with du -sh ~/.ollama/models.
Done with Ollama entirely? Quit the Ollama menu bar app and delete the app from Applications, then use Go to Folder to open ~/.ollama and drag the models folder inside it to the Trash. What stays behind is a few kilobytes of settings and keys, and every model can be re-created by reinstalling and pulling again.

The official way: this tool ships its own cleanup command, which handles locks and indexes correctly.

ollama rm <model>

ollama rm only frees a shared layer once the last model using it is gone, so removing one variant of a model you keep in two sizes may free less than ollama list suggested.

Will it come back?

Only if you invite it back. Ollama never re-downloads a model on its own; the folder grows when you run ollama pull or ollama run for something new. That makes this one of the more satisfying cleanups on a Mac: remove the models you stopped using and the space stays freed. The realistic regrowth pattern is model churn. New releases come out, you pull them to try, and last month's favorites sit there at several gigabytes apiece. A quick ollama list every month or two, deleting anything you do not remember pulling, keeps the folder honest.

Common questions

Can I delete files inside ~/.ollama/models/blobs directly?

Don't. Blobs are content-addressed layers shared between models, and the manifests expect them to exist. Delete a blob by hand and every model that referenced it breaks with an error until you pull it again. ollama rm does the same cleanup correctly, removing only the layers nothing else uses.

Why didn't the folder shrink as much as ollama list said the model was?

Model variants share layers. Two tags of the same model, or a base model and a fine-tune built on it, can point at mostly the same blobs. ollama list reports each model's full logical size, but deleting one only frees the layers no remaining model still references.

Can I move Ollama models to an external drive?

Yes. Recent versions of the Mac app have a setting for where models are stored; point it at a folder on the external drive. If you run Ollama from the terminal instead, set the OLLAMA_MODELS environment variable to that path before starting it (the menu bar app ignores variables set in your shell profile). Move the existing models folder there first if you want to keep what you already have. The catch: those models are unavailable whenever the drive is unplugged.

Does deleting a model remove my chat history or custom prompts?

No. The weights contain no conversation data, so chats in whatever app you use on top of Ollama survive. One thing to watch: models you created yourself with ollama create live in this same store. Keep the Modelfile and you can rebuild them; the base weights they were built from re-download like anything else.