How to delete Ollama models (and list, stop, update and move them)

How do you delete a model in Ollama?

Run ollama rm <model> with the exact name from ollama ls, for example ollama rm llama3.1:8b. Ollama unloads the model if it's running, deletes it, and prints deleted 'llama3.1:8b'. You can name several models in one command. There is no confirmation prompt, so check the name first.

ollama ls                              # find the exact NAME, including the tag
ollama rm llama3.1:8b                  # delete one model
ollama rm llama3.1:8b qwen3:8b         # delete several at once

That's the whole answer for most people. The rest of this page covers the other housekeeping jobs: removing everything, listing, stopping, updating and moving Ollama models.

How to remove all Ollama models at once

Ollama has no built-in "remove all" command: ollama rm needs at least one model name. So you feed it the NAME column of ollama ls. The first line of that output is the header row, which is why both loops below skip it.

# macOS / Linux: delete every downloaded model
ollama ls | awk 'NR>1 {print $1}' | xargs ollama rm

# Windows PowerShell: delete every downloaded model
ollama ls | Select-Object -Skip 1 | ForEach-Object { ollama rm ($_ -split '\s+')[0] }

These loops are ours, built from the documented commands, so run ollama ls first and make sure you really want everything gone. To keep a few, delete by name instead. Re-downloading a model later is one ollama pull.

How to list and check downloaded Ollama models

Run ollama ls, the short alias for ollama list. It prints one row per downloaded model with four columns: NAME, ID, SIZE and MODIFIED. NAME is what you pass to rm, run and show. SIZE is the disk space that model's files take.

ollama ls                   # every downloaded model and its size
ollama show qwen3:8b        # details for one model: parameters, quantization, context, license
ollama cp qwen3:8b my-qwen  # copy a model under a new name (e.g. before customizing it)

The two you'll run most are ollama ls (to see what's taking up space) and ollama rm (to clear out models you tried once and don't use). Since models are gigabytes each, a quick look every so often keeps your disk healthy.

How to stop a running model (or all of them)

Run ollama stop <model> to unload a model from memory right away. Without it, Ollama keeps a model loaded for 5 minutes after its last request, then unloads it. Stopping frees VRAM and RAM. It does not delete anything from disk.

ollama ps shows what's loaded. Its PROCESSOR column also tells you where the model sits: "100% GPU", "100% CPU", or a split like "48%/52% CPU/GPU", which means part of it spilled into system memory and will run slower.

ollama stop takes exactly one model name, so stopping everything is another small loop:

# macOS / Linux: unload every running model
ollama ps | awk 'NR>1 {print $1}' | xargs -n1 ollama stop

# Windows PowerShell: unload every running model
ollama ps | Select-Object -Skip 1 | ForEach-Object { ollama stop ($_ -split '\s+')[0] }

To change the 5-minute default for every model, set the OLLAMA_KEEP_ALIVE environment variable on the server. It takes a duration like "10m" or "24h", a number of seconds, 0 to unload right after each answer, or a negative number to keep models loaded indefinitely. Apps that call the API can override it per request with the keep_alive parameter.

How to update an Ollama model

Run ollama pull <model> again. It fetches the current version of that tag. There is no separate update command. Ollama prunes unused model files when the server starts, unless you set OLLAMA_NOPRUNE, so an update doesn't leave an old copy filling your disk.

This is worth doing for models you rely on, since a re-published tag can include fixes. Updating Ollama itself is a different job. On Windows the app downloads its own updates (Ollama's Windows docs list the folder they land in). On Linux, run the install script again.

Where Ollama stores models, and how to move them

Ollama's FAQ gives the default model folders:

To use a bigger drive, set the OLLAMA_MODELS environment variable to the new folder and restart Ollama. On Windows: quit Ollama from the taskbar, open Settings and search "environment variables", add OLLAMA_MODELS for your account, then start Ollama again. On a Mac running the app, use launchctl setenv OLLAMA_MODELS /path/to/folder and restart the app. On Linux with systemd, add Environment="OLLAMA_MODELS=/path/to/folder" under [Service] via systemctl edit ollama.service, then reload and restart. The Linux service runs as the ollama user, so give it the folder with sudo chown -R ollama:ollama /path/to/folder.

Changing the variable doesn't move files that are already downloaded. Either re-pull the models you want, or move the old folder's contents yourself before restarting.

Deleting the folder by hand is the blunt alternative to ollama rm, and it's easy to get wrong. Use rm for models. Reach for the folder only when you are uninstalling Ollama completely. Ollama's uninstall docs list the folders for each system, and note that on Windows the uninstaller won't remove models you moved with OLLAMA_MODELS.

Deleting a model from code

If you manage Ollama from a script or an app, the API has a delete endpoint. It does the same job as ollama rm for one model:

curl -X DELETE http://localhost:11434/api/delete -d '{"model": "llama3.1:8b"}'

Short on space and wondering what to keep? On a 12GB or 16GB graphics card, a well-chosen mixture-of-experts model often replaces two or three smaller dense ones. That's a good excuse for a cleanup. Use the VRAM calculator to check a model fits before you pull it.

Managing local models comes down to five Ollama commands, ls, rm, ps, stop and pull, plus a bit of disk discipline. Keep your go-to models installed, remove the rest, and size them to your hardware. New to Ollama? Start with the beginner's guide and the Ollama vs LM Studio comparison. Want your models inside an editor? See how to use Ollama with Claude Code.