This downloads the model weights — the numbers that define how the AI behaves — onto your disk. It's a one-time fetch per model (unless you delete it, or Ollama updates the tag).
Use the exact name you wrote down in Lesson 1 — pick your model.
The command
Open your terminal (the same one as before) and run one of these — swap in your own choice if it differs:
8 GB machines:
ollama pull llama3.2
16 GB machines (pick one):
ollama pull llama3.1:8b
ollama pull qwen2.5:7b
24 GB+ machines (pick one):
ollama pull qwen2.5:14b
ollama pull mistral-nemo
What happens
- Ollama shows a download progress bar (percent and MB).
- The first pull can take 5–20 minutes on home broadband — it depends on model size and your connection.
- When it finishes, you'll see a success message and the model name.
You should see: the progress completing without error, and the terminal returning to a prompt when it's done.
If it fails
| Symptom | Try this |
|---|---|
connection refused | Ollama isn't running — redo the confirm step; on Linux, sudo systemctl start ollama |
| Out of disk space | Free up ~5 GB, or pick a smaller model (llama3.2) |
| Very slow, then an error on 8 GB | Model's too large — switch to llama3.2 or phi4-mini |
Go make that brew while it downloads. The first pull always feels slow; pull a second model later the same day and it flies, because you're no longer staring at the bar.
Continue — talk to it.