On a busy workstation, an idle language model can compete with other applications for memory. Ollama can stop the running model to release its system RAM or GPU memory without deleting the downloaded files.
The Ollama server must be reachable, with the target model already loaded; unloading a model leaves the server available and does not quit the Ollama application or stop its service.
If the model reloads immediately, another client may still be sending requests. Keep-alive settings control automatic unloading between requests, while a new request can load an installed model again.
Steps to stop a running Ollama model:
- Identify the loaded model's full name and tag.
$ ollama ps NAME ID SIZE PROCESSOR CONTEXT RUNNER UNTIL qwen2.5:0.5b a8b0c5157701 630 MB 100% CPU 4096 llamacpp 9 minutes from now
The target below is qwen2.5:0.5b; another model requires its own NAME value, including the tag. If no models appear, none are currently loaded on that server. 100% CPU means the model occupies system memory; GPU or mixed placement depends on the hardware.
Related: How to list running Ollama models - Stop the loaded qwen2.5:0.5b model.
Unloading an in-use model can interrupt active requests. On a shared server, other clients may be affected or may load the model again.
$ ollama stop qwen2.5:0.5bA successful stop normally returns without output.
- Confirm that the target model has disappeared from the running list.
$ ollama ps NAME ID SIZE PROCESSOR CONTEXT RUNNER UNTILOnly headings remain when no models are loaded. If other models are running, their rows can remain; the target's absence is what matters. A connection error is not an empty list.
- Verify that the same model is still installed.
$ ollama list NAME ID SIZE MODIFIED granite3.2-vision:2b 3be41a661804 2.4 GB 4 hours ago all-minilm:latest 1b226e2802db 45 MB 5 hours ago qwen3:0.6b 7df6b6e09427 522 MB 5 hours ago qwen2.5:0.5b a8b0c5157701 397 MB 5 hours ago
The qwen2.5:0.5b row remains on disk despite its absence from the running list. The next request can use those downloaded files without pulling the model again.
Related: How to list Ollama models