On Thursday, enterprise AI firm Cohere introduced its inaugural voice model: Transcribe. This open-source automatic speech recognition tool is designed for applications such as note-taking and speech analysis. With a relatively compact size of 2 billion parameters, the model is optimized to run on consumer-grade GPUs, enabling users to host it themselves.

EMBED_PLACEHOLDER_0

Currently, Transcribe supports 14 languages: English, French, German, Italian, Spanish, Portuguese, Greek, Dutch, Polish, Chinese, Japanese, Korean, Vietnamese, and Arabic. According to Cohere, Transcribe outperforms competing models like Zoom Scribe v1, IBM Granite 4.0 1B, ElevenLabs Scribe v2, and Qwen3-ASR-1.7B Speech on the Hugging Face Open ASR leaderboard. It achieves an average word error rate (WER) of 5.42, which is lower than any other model on this benchmark.

The company also reports that human evaluators rated Transcribe’s transcriptions for accuracy, coherence, and usability, giving it an average win rate of 61% over other models. However, the model did lag behind competitors when transcribing Portuguese, German, and Spanish.

EMBED_PLACEHOLDER_1

Cohere states that Transcribe can process 525 minutes of audio in just one minute, a notably high throughput for its model class. The firm plans to integrate Transcribe into its enterprise agent orchestration platform, North, and is offering the model for free via its API. Additionally, it will be accessible on Model Vault, Cohere’s managed inference platform.

Speech recognition models are gaining traction as demand rises for note-taking and dictation applications such as Granola and Wispr Flow. Earlier this year, Cohere reportedly informed investors that it was on track to generate $240 million in annual recurring revenue by 2025. CEO Aidan Gomez was quoted suggesting that the startup could go public “soon.”