Models

Detexts resolves your selection to an on-device or cloud transcriber. Download and switch models anytime in Settings → Transcription Models.

Whisper

On-device via whisper.cpp with Metal acceleration.

• Tiny — smallest and fastest, lower accuracy.
• Small — a balanced option.
• Turbo — the most accurate. Recommended.

NVIDIA Parakeet

CoreML / Neural Engine. The fastest local ASR, English only — ideal if you dictate in English and want minimal latency.

Cloud (optional)

GPT-4o mini Transcribe with your own OpenAI key, for the widest language coverage. Audio leaves the device only when you choose a cloud model.