Skip to content

Install a model

AI models are optional downloads. Everyday listening and audio settings work without them.

Model in Settings Used for Download
Speech model Create timed transcripts About 600 MB
Language model · Quality Summaries and Ask 2.7 GB
Language model · Fast Summaries and Ask, with lower memory use 1.3 GB

Publisher transcripts need no speech model. Summaries and Ask always need a language model.

Open Settings → Local AI, then select Download. Studio and Ask also offer download cards. Files are verified before use; no restart is needed.

Quality is preferred when installed. Select Use beside another installed language model to switch. Only one stays loaded at a time.

Use Check files for an unverified installation or Verify files to recheck an installed model.

  • On-device AI controls whether AI features are shown; it does not delete model files.
  • Transcribe new downloads is off by default and skips episodes with timed publisher transcripts.
  • Summarize after transcribing queues a summary when a language model is available.

Downloads come from Hugging Face. See model troubleshooting.