Delete Unused Ollama Models and Free Mac Storage

List your downloaded models with ollama ls, then remove a model you no longer need with ollama rm and its exact name. Stopping a running model frees its loaded memory; it is not the same operation as deleting its downloaded files.

Local model experiments can leave several large downloads behind. Before deleting anything, decide which model your actual workflow uses and which ones were one-off comparisons.

Inventory models and current activity

ollama ls
ollama ps

The first command lists available models; the second shows loaded models. Compare the names with configuration in any local chat interface or development tool that connects to Ollama. A model you have not opened directly may still be used by another application.

The official Ollama CLI reference documents model listing and removal. Write down the complete model name and tag before changing anything, especially if you have several variants with similar names.

Remove one selected model

Stop using the model and finish any active requests. Here is an example for a model named llama3.2; substitute the exact unused name from your own list:

ollama stop llama3.2
ollama rm llama3.2
ollama ls

Do not run the example merely because the name looks familiar. The decision is whether you can do without that specific download. If it is a custom model, preserve the source files and configuration needed to recreate it before removal.

Find the disk storage you are measuring

Ollama documents the default macOS model location as ~/.ollama/models, with OLLAMA_MODELS available to override it. Its FAQ also explains loaded-model behavior and configuration. Check the active setup if the default directory is missing or unexpectedly small.

Measure the directory and the volume’s free space before and after removal. Do not promise yourself savings by adding every displayed model size: related model entries and filesystem behavior can make apparent totals misleading. Let the actual measurement determine the result.

Keep the surrounding application data intact

A whole-folder deletion is broader than removing one model. It makes it harder to distinguish lost configuration from a missing model and can disrupt a setup you intended to keep. Use the model manager for model removal and leave unrelated application state alone.

If disk usage is still high, check whether you also downloaded the same family through LM Studio or the Hugging Face cache. Cleaning Ollama does not automatically clean those separate stores.

Make your next model download intentional

Keep a small inventory with each retained model’s purpose: daily assistant, coding comparison, embedding workflow, or offline fallback. Before downloading another variant, choose what question the experiment will answer and when you will retire it.

After cleanup, run one normal request through the application you actually use. If it asks to download a removed model, update that application’s selection or restore the required model. A tidy model list is useful only if your working setup still points to something available.