Skip to content

Model Management

Manage everything in Settings → Models.

Transcription engines

  • SenseVoice (250 MB): bundled with the installer and seeded on first launch; re-downloadable from Settings after deletion;
  • Whisper (77 MB – 1.6 GB): five on-demand tiers — tiny / base / small / medium / large-v3-turbo. Existing GGUFs from LM Studio or Jan can be symlinked (zero extra disk).

AI translation models (Qwen3)

TierSizeRecommended RAM
4B2.5 GB8 GB
8B5.0 GB16 GB
14B9.0 GB32 GB

Settings recommends a tier based on your RAM. Upgraders from Qwen 2.5 can clean up old model files from Settings in one click.

Runtime status

Settings shows transcription model runtime and AI model runtime separately. Transcription covers SenseVoice / Whisper; AI covers Qwen3 translation, transcript polish and AI summaries.

Running means the model is actively processing work. Loaded means the model is still in memory and will auto-release after being idle. This helps explain whether fan noise, memory use or an active task is coming from Subcast.

Storage location

~/Library/Application Support/Subcast/models/ (macOS). Deleting a model never affects completed transcripts or translation caches.

Released under the Apache-2.0 License