The add-in does not ship with a model. It lists whatever Ollama has installed on your machine, and you pick one at the top of the task pane. This page is about which one to pick and how to get more.
Fast picks #
Match the model to the RAM you have free, not to the RAM you own. As a rule of thumb a model needs roughly its download size in free memory, plus room for whatever you send it.
| Your machine | Start with | What to expect |
|---|---|---|
| 8 GB RAM, no GPU | gemma3:1b |
Quickest to answer. Fine for short summaries and tidying text, weaker on reasoning. |
| 16 GB RAM (most laptops) | gemma3:4b |
The default recommendation. Good quality at a speed you will not mind. |
| 16 GB RAM and a decent GPU | llama3.1:8b or qwen3:8b |
Noticeably better on reasoning and longer instructions. |
| 32 GB RAM or more | gemma3:12b and up |
Best answers this side of a cloud model. Slow without a GPU. |
If answers feel sluggish, drop one row rather than living with it. See answers are very slow for the other things that help.
Install another model #
Models come from Ollama, not from Excel, so you pull them in a terminal:
ollama pull gemma3:4b
- The download runs once and the model then lives on your disk.
- Reopen the task pane afterwards and the new model appears in the dropdown.
ollama listshows what you already have,ollama rm <model>removes one you are done with.
Browse the full library #
Every available model, with its exact download size and what it is built for, is at ollama.com/search. Two things to watch for when you pick from there:
- The number after the colon is the parameter count, and it drives both quality and speed.
:4band:8bare the useful range for most machines. - Embedding models cannot chat. If a model page says it is for embeddings, it will not work in the pane or in an
=XLLAMAformula.
The first answer after switching models is always slower: Ollama has to load it into memory first. Every answer after that runs at the model’s normal speed.
