Model memory can remain occupied after an Ollama request finishes. The running model list identifies which models are still loaded, along with their memory use and processor placement.
Loaded models are separate from downloaded models. An installed model can stay on disk while absent from the running list, and a resident model can be idle between requests.
An Ollama server must be running and reachable from your client. The list describes that server's memory at the time of the check; automatic expiry or another client's request can change it moments later.
Steps to list running Ollama models:
- List the models currently loaded by the Ollama server.
$ ollama ps NAME ID SIZE PROCESSOR CONTEXT RUNNER UNTIL qwen2.5:0.5b a8b0c5157701 630 MB 100% CPU 4096 llamacpp 4 minutes from now
The NAME and ID columns identify the model with its tag and shortened digest. SIZE describes the loaded model's memory footprint rather than its download size. PROCESSOR shows memory placement: 100% CPU means system memory, 100% GPU means GPU memory, and a split percentage means both. This is not a processor utilization reading. CONTEXT is the allocated context length in tokens, RUNNER identifies the inference backend, and UNTIL is the remaining time before idle unloading.
Related: How to check Ollama GPU supportA server with no loaded models returns only the headings, even when models are installed. For a populated list, a model must have received a request and still be resident.
Related: How to run a model in Ollama$ ollama ps NAME ID SIZE PROCESSOR CONTEXT RUNNER UNTIL - Confirm the loaded model details through the same server's API.
$ curl -sS http://localhost:11434/api/ps {"models":[{"name":"qwen2.5:0.5b","model":"qwen2.5:0.5b","size":630707976,"digest":"a8b0c51577010a279d933d14c2a8ab4b268079d44c5c8830c0a93900f1827c67","details":{"parent_model":"","format":"gguf","family":"qwen2","families":["qwen2"],"parameter_size":"494.03M","quantization_level":"Q4_K_M"},"expires_at":"2026-10-28T04:55:33.445780669Z","size_vram":0,"context_length":4096,"runner":"llamacpp"}]}
The models array lists resident models. Each entry contains the full digest, size in bytes, size_vram in GPU memory bytes, context_length in tokens, and expires_at as an expiry timestamp. The model above has zero GPU memory and matches the CLI's 100% CPU placement. The API has no processor string field. Values vary by model, hardware and request settings.
An empty runtime returns the response below. A connection error or error response is different from an empty list. If the CLI uses another server through OLLAMA_HOST, the API URL must address that same server.
$ curl -sS http://localhost:11434/api/ps {"models":[]}