> ## Documentation Index
> Fetch the complete documentation index at: https://elephas.app/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Which Local AI Model Should You Choose?

> Pick the right Local AI model in Elephas for your computer's memory. Compare Qwen 2B, 4B and 9B on speed and quality, switch models, and manage the downloads.

Most people should leave the model on **Choose for me**. Elephas then picks the best model your computer can run comfortably. Choose by hand if you want faster answers on a busy computer, or better answers on one with memory to spare.

## Before you start

* Read [Run Local AI](/docs/ai-options/run-local-ai) to set it up.
* Know roughly how much memory (RAM) your computer has. On Mac, choose **Apple menu › About This Mac**. On Windows, open **Settings › System › About**.

## The models

| Model | Download | Suggested for | In the app's words |
| - | - | - | - |
| **Qwen 2B** | About 1.2 GB | Any computer | "Fastest, and the lightest on memory. Answers are shorter and simpler." |
| **Qwen 4B** | About 2.6 GB | 8 GB of memory or more | "A good balance of speed and quality on most computers." |
| **Qwen 9B** | About 5.3 GB | 20 GB of memory or more | "The best answers, if the computer has the memory to spare." |

The download also includes a small AI engine the first time. Elephas shows the exact size before it downloads anything.

## Steps

### See which model fits your computer

<Steps>
  <Step title="Open the Local AI card">
    Open **Settings › AI** (⌘, on Mac, Ctrl+, on Windows). Click **Local (on this computer)**. If you aren't ready to download, click **Not now**. The card stays open.
  </Step>

  <Step title="Open Which model">
    Click **Which model**. The list starts with **Choose for me**, then each model with its size. The one that suits your computer says **Recommended for this computer.** A model that is likely too big says **Probably too big for this computer. You can still try it.**
  </Step>

  <Step title="Pick a model">
    Click **Choose for me** or a model. Picking downloads nothing by itself: "Changing this downloads nothing now. It applies when you next set up Local AI."
  </Step>
</Steps>

### Switch to another model

<Steps>
  <Step title="Pick the new model">
    In **Settings › AI**, under **Local (on this computer)**, open **Which model** and click the model you want.
  </Step>

  <Step title="Download it">
    The status line says **You chose** followed by the model, **which isn't downloaded yet.** The button below names the download, for example **Download Qwen 9B · 5.3 GB**. Click it.

    If Local AI isn't set up yet, click **Local (on this computer)** again instead. The download dialog names the model you picked.
  </Step>

  <Step title="Wait for it to start">
    Elephas stops the current model, then downloads, checks and starts the new one. Local AI can't answer in the meantime, so switch when you can spare the wait, or use another option for a while. When the status reads **Local AI ready** with the new model's name, Ask uses it.
  </Step>
</Steps>

The steps are the same on Mac and Windows.

## How it works

### How Choose for me decides

Elephas suggests the largest model your computer's total memory qualifies for. On Windows, it also looks at how much memory is free right now and steps down to a smaller model if there isn't enough headroom. If even the smallest model doesn't fit, Local AI shows **Not enough free memory on this computer.**

If Elephas later finds a different model would suit your computer better, the status line adds **A different model suits this computer better.** Click **Set up again** to switch. Elephas never replaces an installed model without your click.

### Speed versus quality

* **Smaller is faster.** Qwen 2B starts quickly and answers fastest, with shorter, simpler answers. Good for quick lookups on any computer.
* **Bigger is better, and slower.** Qwen 9B writes the most careful answers but needs the most memory, and other apps may slow down while it runs.
* **You can check.** Open **Details** under the Local AI card to see numbers from the last answer: how long the model took to load, how long until the first word, how fast it wrote and how much memory it used. It never shows any part of your question or answer.

All three models answer in a single pass from the passages Elephas finds. For long summaries across many documents, cloud AI reads further. See [Choose how Elephas answers](/docs/ai-options/choose-how-elephas-answers).

### Removing a model you no longer use

Elephas doesn't have a button to delete a downloaded model yet. When you switch models, the previous one stays on your computer. To remove every model, along with the rest of Elephas's data, see [Uninstall Elephas](/docs/getting-started/uninstall-elephas).

## Troubleshooting

<AccordionGroup>
  <Accordion title="The model I picked is slow or makes my computer sluggish">
    It may be too big for the memory you have free. Pick **Choose for me** or a smaller model, then download it. Closing other large apps also helps. See [Local AI troubleshooting](/docs/troubleshooting/local-ai-troubleshooting).
  </Accordion>

  <Accordion title="I picked a model but nothing happened">
    Picking only records your choice. Click the **Download** button that names the model, or **Set up again**.
  </Accordion>

  <Accordion title="Not enough free disk space">
    The card says how much to free. Free the space and click **Retry**, or pick a smaller model.
  </Accordion>
</AccordionGroup>

## Related articles

<CardGroup cols={2}>
  <Card title="Run Local AI" href="/docs/ai-options/run-local-ai" />

  <Card title="System requirements" href="/docs/getting-started/system-requirements" />

  <Card title="Local AI troubleshooting" href="/docs/troubleshooting/local-ai-troubleshooting" />

  <Card title="Choose how Elephas answers" href="/docs/ai-options/choose-how-elephas-answers" />
</CardGroup>


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.