Changelog

Everything that ships in Voxpen.

v1.4.0

NewJuly 25, 2026
  • NVIDIA Parakeet TDT v3 engine — up to 110x realtime transcription on the Neural Engine, roughly 10x faster than Whisper large
  • Whisper large-v3-turbo added to the model lineup — 99 languages at 8x the speed of large-v3
  • New Models tab in Settings — download, switch, and manage all seven local models
  • Selecting a model now hot-swaps the engine instantly, no restart needed
  • Context-aware number formatting — 'one of the greatest' stays words, 'twenty percent' becomes 20%

v1.3.0

March 22, 2026
  • Voxpen is now completely free — no trial, no subscription
  • Streamlined, more compact onboarding
  • Transcription pipeline diagnostics and an engine-refresh safety net for consistent quality over long sessions

v1.2.0

February 2026
  • Fixed transcription quality degrading after repeated use
  • Fixed long recordings skipping sentences
  • Fixed intermittent 'no speech detected' failures and silent recording failures
  • Fixed dashboard history not appearing, plus dashboard performance improvements

v1.0.1

February 2026
  • Fixed microphone permission not registering in System Settings
  • Fixed lingering Accessibility permission dialogs
  • Onboarding now re-appears if permissions are missing after a reinstall

v1.0.0

February 19, 2026
  • Initial release — private, on-device dictation for macOS with global hotkey, auto-paste, and Polish Mode