Cohere launches Transcribe, an open-source ASR model that tops accuracy benchmarks

Cohere Enters Voice AI With Dedicated ASR Model

Enterprise AI company Cohere has launched Transcribe, marking its first venture into voice technology with an open-source voice model specifically built for transcription tasks. The automatic speech recognition model is designed to handle real-world applications including note-taking, meeting transcriptions, and speech analysis for customer support analytics1

1

. At just 2 billion parameters, Transcribe is relatively lightweight and optimized for deployment on consumer-grade GPUs, making it accessible for organizations that want to self-host their speech recognition models1

1

.

Benchmark Performance and Multilingual Support

Transcribe has achieved top ranking on the Hugging Face Open ASR Leaderboard with an average word error rate of 5.42%, outperforming competing models including Zoom Scribe v1, IBM Granite 4.0 1B, ElevenLabs Scribe v2, Qwen3-ASR-1.7B Speech, and Whisper Large v31

1

2

2

. The model currently supports 14 languages including English, French, German, Italian, Spanish, Portuguese, Greek, Dutch, Polish, Chinese, Japanese, Korean, Vietnamese and Arabic1

1

. When evaluated by human assessors for accuracy, coherence and usability, Transcribe achieved an average win rate of 61% over rival models, though it showed weaker performance on Portuguese, German and Spanish transcription tasks1

1

.

Processing Speed and Enterprise Integration

Cohere reports that Transcribe can process 525 minutes of audio in just one minute, delivering high throughput for its model class1

1

. The company is making the model available through multiple channels: free access via its API for experimentation, download as open source on Hugging Face, and production deployment through Model Vault, Cohere's managed inference platform1

1

2

2

. Cohere plans to integrate Transcribe into North, its enterprise agent orchestration platform, expanding capabilities beyond basic transcription2

2

.

Strategic Positioning in Growing Voice AI Market

The launch positions Cohere to compete in the rapidly expanding market for speech recognition models, driven by surging demand for dictation and note-taking applications like Granola and Wispr Flow1

1

. Founded in 2019 in Toronto, Cohere specializes in LLMs and generative AI for the enterprise market, with particular emphasis on data privacy, security and customizability2

2

. The company reportedly told investors it was generating annual recurring revenue of $240 million in 2025, and CEO Aidan Gomez has indicated the startup may go public soon1

1

. This week, Cohere also announced a strategic partnership with Saab to advance AI technologies within the aerospace sector, marking expansion beyond its core enterprise focus2

2

.