Why is the first AI load slow?
The first load is slow because your browser has to download the AI model before it can run. The model is large, so the download can take several minutes depending on your connection.
What's happening
- When you turn on the AI, the Research Assistant downloads the model weights. The status shows "Downloading Model" with a percentage.
- After the download, the model is loaded onto your graphics chip. The status shows "Initializing."
- When it's ready, the AI features become available.
The default model is Gemma 2 (2 billion parameters). The app's model list gives its download size as about 1.4 GB.
Later loads
The model is cached in your browser, so later loads skip most of the download and are faster. After the first download, the model can also run offline.
Tips
- Use a fast, unmetered connection for the first download.
- Keep the tab open until loading finishes.
- If loading fails, use the retry option. Your projects aren't affected.
- Clearing your browser's site data removes the cached model, so the next load downloads it again.