Superwhisper adds Cohere Transcribe for local dictation with 5.35% WER

aidangomez · x · 2026-07-29

Superwhisper has partnered with Cohere to add Transcribe to its model library for local dictation.

The post says the 2B-parameter model is one of the top-ranked ASR models on Hugging Face’s Open ASR leaderboard, with a 5.35% average WER that drops to 1.25% on clean audio. It is described as beating Whisper Large v3 on accuracy and throughput, supporting 14 languages, and running on-device under Apache 2.0.

Original post →

More from Apps

Apps channel →