Cohere Transcribe Released

Cohere has entered the competitive speech-to-text arena with the release of Transcribe, a 2-billion parameter model designed to be a top-tier open-source option. Released under the business-friendly Apache 2.0 license, the model is immediately available for commercial and research use without restrictive fees. Cohere claims the model achieves state-of-the-art (SOTA) performance among openly available transcription models, positioning it as a direct challenger to both other open-source projects and closed commercial APIs. This strategic move leverages Cohere's established reputation in multilingual AI, built on models like their highly-regarded Aya translation model.

The model's key feature is its broad multilingual support, covering 14 languages across three major regions: European languages (English, French, German, Italian, Spanish, Portuguese, Greek, Dutch, Polish), Asian languages (Chinese, Japanese, Korean, Vietnamese), and Arabic from the MENA region. This wide coverage addresses a significant gap in the open-source ecosystem, where high-quality, multilingual transcription has often been lacking. By providing a capable, freely licensed model, Cohere is empowering developers to build voice-enabled applications—from meeting assistants to content subtitling tools—without vendor lock-in or per-minute costs. The release signals a continued push by major AI labs to open-source core technologies, increasing accessibility and fostering innovation in the speech AI domain.