Cohere launches open-source voice model Transcribe, pushes deeper into enterprise AI stack

New updates have been reported about Cohere.

Cohere has introduced Transcribe, its first voice model and an open-source automatic speech recognition system, positioning the company to capture more enterprise spend in multimodal AI workflows. The 2-billion-parameter model is optimized for consumer-grade GPUs, supports 14 major languages, and is designed for customers that want to self-host or tightly control data, reinforcing Cohere’s focus on privacy- and compliance-sensitive use cases.

On industry benchmarks, Cohere reports that Transcribe tops the Hugging Face Open ASR leaderboard with a 5.42 average word error rate, outperforming several notable rivals and winning 61% of pairwise human evaluations on accuracy, coherence, and usability, though it trails some competitors in Portuguese, German, and Spanish. The company plans to integrate Transcribe into its North enterprise agent orchestration platform and distribute it both via API at no charge and on its managed Model Vault service, broadening its platform lock-in potential as demand accelerates for speech-enabled tools like note-taking and dictation apps; this move follows disclosures to investors that Cohere is targeting $240 million in annual recurring revenue in 2025 and that CEO Aidan Gomez is considering a near-term IPO, suggesting voice capabilities could become a meaningful contributor to product differentiation and valuation.