Keep an Ollama Model Loaded (or Unload It Now): keep_alive, Preloading and ollama stop
Note: Some future links on this site may be affiliate links. We never claim hands-on testing; guides are researched from the public sources listed in each article.
How we wrote this: A plain-English summary of the official Ollama FAQ, checked October 29, 2026.
The first request to a model is slow because Ollama has to load it into memory. After that, Ollama keeps it loaded for a while. These settings control that trade-off between speed and free RAM or VRAM.
The default: 5 minutes
The Ollama FAQ says models stay in memory for 5 minutes after the last request, then unload. Run ollama ps to see what's loaded and the UNTIL column showing when each model will unload.
Unload a model right now
ollama stop llama3.2
That's handy before a game, a video call or loading a bigger model.
Preload a model
Send an empty request and the model loads without producing any text:
curl http://localhost:11434/api/generate -d '{"model": "mistral"}'
It works with /api/chat too. From the CLI: ollama run llama3.2 "".
The keep_alive parameter
On /api/generate and /api/chat, keep_alive sets how long the model stays loaded after that request:
| Value | Effect |
|---|---|
"10m", "24h" |
A duration string |
3600 |
A number of seconds |
-1 (or any negative number) |
Keep it loaded indefinitely |
0 |
Unload immediately after the response |
Keep a model loaded:
curl http://localhost:11434/api/generate -d '{"model": "llama3.2", "keep_alive": -1}'
Free the memory:
curl http://localhost:11434/api/generate -d '{"model": "llama3.2", "keep_alive": 0}'
Change the default for every model
Set OLLAMA_KEEP_ALIVE (same value types) on the server and restart it. A keep_alive value in an API request overrides the server setting. How you set server variables depends on your system: launchctl on a Mac, systemctl edit ollama.service on Linux, Environment Variables in Windows Settings. See Ollama on Mac, Linux and Windows.
Which to choose
- Shared or low-memory machine: keep the default, or use
0for one-off jobs. - A local app or coding assistant you use all day:
-1or a few hours avoids reload delays.
Related: running several models at once, is Ollama using my GPU?, CLI cheat sheet.