Cohere Transcribe hits #1 on ASR leaderboard
Cohere's 2B parameter open-source speech model outperforms OpenAI's Whisper Large v3 in accuracy and speed, delivering a 27% improvement in Word Error Rate. Optimized for enterprise environments, it supports 14 business-critical languages under an Apache 2.0 license for local or cloud deployment.
Cohere Transcribe is a precision instrument designed to dismantle Whisper’s dominance in the enterprise transcription market by trading breadth for extreme performance. It achieves a 5.42% Word Error Rate, unseating Whisper Large v3 from the top of the Hugging Face Open ASR Leaderboard. Its asymmetric architecture (heavy encoder, light decoder) prioritizes throughput, processing 8 hours of audio in just one minute on standard hardware. Optimized for noisy boardroom acoustics and diverse accents, it is ideal for privacy-focused local deployments, though it lacks Whisper's extensive language support and auto-detection.
DISCOVERED
11d ago
2026-03-31
PUBLISHED
11d ago
2026-03-31
RELEVANCE
AUTHOR
gmays