On-device AI models
Choose a compatible model and manage local storage.
Checks whether this device can run AI in the browser, then downloads, switches or deletes the on-device model that the AI tools use.
How to use it
- Select Check this device. superPDF shows whether your browser can run AI on this device, how much memory and storage it reports, and which model suits it.
- Select Download next to a model. The download shows its progress, and you can cancel it; files that finished downloading are kept.
- Open an AI tool. It uses the model marked In use. Select Use this model to switch, or Delete to remove one.
Models
| Model | Download | Best for |
|---|---|---|
| Gemma 4 E2B (Google) | 2.0 GB | Computers with 8 GB of memory or more. The default: the best answers, summaries and translations, in many languages. |
| Qwen3 0.6B (Qwen) | 606 MB | A fallback for computers with 4 GB of memory. Answers are simpler and miss details and citations more often. |
The sizes include the software that runs the models. See Privacy for what is downloaded and from where.
Requirements
A current Chrome or Edge on a computer, or Safari 26 or later, and enough free storage for the download. On a recent computer, a tool loads Gemma 4 E2B in a few seconds and answers in one to two seconds; computers with less powerful graphics are slower.
Voice and transcription models
On-device AI models also lists the voice for PDF to audio and the speech models for Transcribe audio to text, with Download and Delete. The voice and the Standard speech model come in a faster version that uses the graphics processor (338 MB and 209 MB) and a smaller one that runs on any device (105 MB and 80 MB). The Accurate speech model (252 MB) has only the smaller kind. Each tool also offers its download the first time you use it.
Privacy
Runs in your browser. Your files are processed on your device and never uploaded.