Cohere AI Releases Cohere Transcribe: A SOTA Automatic Speech Recognition (ASR) Model Powering Enterprise Speech Intelligence

Cohere released Cohere Transcribe the company announced a new automatic speech recognition model.

Cohere Transcribe the model uses a Conformer encoder and a lightweight Transformer decoder hybrid architecture with a large Conformer encoder paired with a lightweight Transformer decoder.

Training procedure the model was trained with supervised cross-entropy training objective minimized difference between predicted text and ground-truth transcripts.

Cohere Transcribe the model supports 14 languages includes English, German, French, Italian, Spanish, Portuguese, Greek, Dutch, Polish, Arabic, Vietnamese, Chinese, Japanese, and Korean.

Benchmark result the model ranked #1 on the Hugging Face Open ASR Leaderboard on 2026-03-26 with average WER 5.42% average word error rate across benchmarks including AMI, Earnings22, GigaSpeech, LibriSpeech, SPGISpeech, TED-LIUM, VoxPopuli was 5.42%.