Cohere launches an open-source voice model specifically for transcription

Enterprise AI institution Cohere connected Thursday launched its archetypal dependable model: Transcribe is an open-source automatic code designation exemplary that tin beryllium utilized for tasks similar note-taking and code analysis.

Relatively airy astatine conscionable 2 cardinal parameters, the exemplary is meant for usage with consumer-grade GPUs for those who privation to self-host it. It presently supports 14 languages: English, French, German, Italian, Spanish, Portuguese, Greek, Dutch, Polish, Chinese, Japanese, Korean, Vietnamese and Arabic.

Cohere says Transcribe beats models specified arsenic Zoom Scribe v1, IBM Granite 4.0 1B, ElevenLabs Scribe v2, and Qwen3-ASR-1.7B Speech connected the Hugging Face Open ASR leaderboard, achieving an mean connection mistake complaint (WER) of 5.42, little than immoderate different exemplary connected the benchmark.

The institution claims Transcribe had an mean triumph complaint of 61% implicit different models erstwhile quality evaluators assessed its transcriptions for accuracy, coherence and usability. However, the exemplary fell down its rivals erstwhile it had to transcribe Portuguese, German and Spanish.

Cohere says Transcribe tin process 525 minutes of audio successful a minute, which is precocious for its people of model.

The institution is readying to integrate Transcribe into its endeavor cause orchestration platform, North, and is making the exemplary disposable done its API for free. The exemplary volition besides beryllium disposable connected Model Valut, Cohere’s managed inference platform.

Speech designation models are increasing progressively fashionable arsenic request grows for note-taking and dictation apps similar Granola and Wispr Flow.

Earlier this year, Cohere reportedly told investors that it was generating yearly recurring gross of $240 cardinal successful 2025, and its CEO, Aidan Gomez, was cited arsenic saying that the startup may spell nationalist “soon”.