Introducing: Cohere Transcribe – a new state-of-the-art in open source speech recognition.

This is Cohere’s first speech-to-text model release and a huge step towards our goal of delivering enterprise speech intelligence into North, Cohere’s agentic AI orchestration platform.

“We’re genuinely impressed with what Cohere has built with Transcribe. The speed is exceptional - turning minutes of audio into usable transcripts in seconds - and it immediately unlocks new possibilities for real-time products and workflows,” said Paige Dickie, VP, Radical Ventures.

Our open-source speech-to-text model has secured the top spot for English language accuracy on HuggingFace’s Open ASR model leaderboard, achieving an impressive word error rate of just 5.42% and validated by human evaluation. We've also successfully achieved one of the strongest accuracy-speed ratios among speech models of a comparable size.

Learn more: Download Cohere Transcribe:

Apache 2.0! 👏

Speech-to-text finally getting real competition is good for everyone building voice features.

👀

How is Parakeet not being compared anywhere here? That is by far the best STT model I have used

Open-source Whisper has been the fallback for teams that can't send audio to OpenAI. If Cohere Transcribe actually beats it on accuracy, that flips the default for regulated industries running local inference.

Finally an open-source ASR model that actually beats Whisper Large v3 on the leaderboard (5.42% WER 🔥). 2B params, runs fully locally via HF Transformers, Apache 2.0. This is huge for privacy-focused teams and on-prem deployments. Great work Cohere!"

Cohere Transcribe looks next-level! Can’t wait

Been dabbling with this, results are amazing! Great job @cohere

been using this for some customer calls at ElderVoice and the accuracy on older voices is surprisingly good. most models struggle there

@Scobleizer Put this in smart glasses and you’d be helping the hard hearing tremendously

@UnslothAI GGUF when?

@grok how good is this vs OpenAI whisper model?

The color scheme on the huggingface space isn't very legible.

@zachtratar this may or may not be relevant news for you & meeting notes!

conformer-based with 2B params and 5.42% WER on open ASR leaderboard is no joke. the enterprise speech game just got a lot more interesting.

@Scobleizer Dang elevenlabs getting a run for its money! Love the competition

oh my days.. big

Love it. Just made a whisperX style interface for it with VAD and word alignment.

wow, this is awesome

How does this compare with Nvidia Parakeet V3

Can it beat eleven labs?

@AWS will this be available in bedrock?

Open source speech recognition hitting state-of-the-art is a huge win for the developer community. Cohere continues to deliver where enterprise needs it most

Impressive

@llmgram

Urdu????????????????????

The race for the ultimate #SpeechRecognition model is heating up. With Cohere Transcribe entering the open-source arena, the ecosystem is no longer reliant on just one or two major players for state-of-the-art. Incredible release by the @cohere team! 🌐🛠️ #ArtificialIntelligence

open source speech recognition getting this competitive is a win for everyone building voice products. cohere making a serious run at whisper is worth paying attention to

Thanks for this banger! Do you think it'll be easy to use it in a streaming setup? Or it was specifically trained for offline transcription.

@grok can this run locally on an android device

open source speech models finally catching up to the hype

Ya’ll cheaper than elevanlabs?