Cohere Launches Open-Source Voice Model – Transcribe
Automatic Speech Recognition
AI Model
Voice AI
Transcribe
Cohere
Open Source
Hugging Face Leaderboard
5
Steady Progression
Media Hype
6/10
Real Impact
5/10
What is the Viqus Verdict?
We evaluate each news story based on its real impact versus its media hype to offer a clear and objective perspective.
AI Analysis:
Moderate media buzz surrounds the release of an open-source voice model, but the technical advancements are incremental. The core AI landscape isn’t experiencing a disruptive shift, limiting the broader industry implications.
Article Summary
Cohere unveiled Transcribe, its first voice model, on Thursday. This open-source Automatic Speech Recognition (ASR) model is targeted at users with consumer-grade GPUs, allowing for self-hosting. Transcribe supports 14 languages, including English, French, German, and Spanish, and achieves a WER of 5.42 on the Hugging Face Open ASR leaderboard, outperforming models like Zoom Scribe and Qwen3-ASR-1.7B. Cohere claims a 61% win rate in accuracy tests against other models. Despite competitive performance, the model showed some weaknesses in Portuguese, German and Spanish transcription. The company intends to integrate Transcribe into its North agent orchestration platform and offer it via API and its managed inference platform, Model Valut. This release aligns with the growing popularity of speech recognition models for applications like note-taking and dictation, mirroring the rise of tools like Granola and Wispr Flow. Cohere’s financial growth, with reported $240M ARR in 2025, adds another layer to the significance of this launch.Key Points
- Cohere launched Transcribe, its first open-source voice model.
- Transcribe supports 14 languages and achieves a competitive WER on the Hugging Face Open ASR leaderboard.
- The model will be integrated into Cohere’s North platform and offered via API and its managed inference platform.