Cohere Transcribe
launchlate March 2026
powered byCohere Transcribe 2B
goblin vibe check:
fast and accurate transcription that you can run yourself or hit via api without worrying about vendor lock-in
open-source enterprise asr model from cohere optimized for accurate multilingual speech-to-text in meetings, support, and analytics workflows.
speed
525
min/min
#1 on the Hugging Face Open ASR leaderboard14-language multilingual transcriptionRuns on consumer GPUs and edge devicesHigh-fidelity transcription for enterprise audio
key features
#1 on the Hugging Face Open ASR leaderboard14-language multilingual transcriptionRuns on consumer GPUs and edge devicesHigh-fidelity transcription for enterprise audio
spec & usage
2B parameter encoder-decoder ASR model released under Apache 2.0 in March 2026
Best fit for meeting notes, customer support QA, and production batch transcription
Available both for self-hosting and through Cohere Model Vault
limitations
No timestamps or speaker diarization
Needs the language specified up front for best accuracy
Works best when paired with VAD because it may transcribe non-speech sounds
scope:
audiovoiceresearchapilocalopen-sourcebenchmark-strongfree
launchlate March 2026