Cohere has officially launched “Cohere Transcribe Arabic,” a high-performance, open-source automatic speech recognition (ASR) model specifically engineered to address the complexities of the Arabic language in professional environments. Built upon Cohere’s 2B frontier ASR model, this release is designed to capture the linguistic diversity inherent in Arabic, including dialectal variations, regional pronunciations, and the common practice of code-switching between Arabic and English. In a significant technological milestone, Cohere Transcribe Arabic achieves the lowest word error rate (WER) on the Hugging Face Arabic ASR Leaderboard, outperforming established alternatives such as OpenAI’s Whisper Large V3 and Meta’s OmniASR.
Also Read: RWS Unveils Trados Studio 2026: Revolutionizing Language Tech via Context-Aware AI and Next-Gen Architecture
Apart from accuracy, the model is designed with enterprise-scale applications in mind, scoring an RTFx rate of 525 that easily beats out any competing solution. Human assessments reinforce this point as well, showing that native speakers prefer the model from Cohere in 95.8 percent of cases compared to Whisper due to its fidelity and use of mixed-language workplace terminology. The company emphasizes that “Transcribe Arabic achieves the highest accuracy for Arabic-language transcription among open-weights models. It is designed to capture the nuances and dialectical richness of Arabic, while being fit for enterprise speech applications.” Now available under an Apache 2.0 license, developers can access the model via Hugging Face, the Cohere API, or Model Vault, marking a substantial advancement in sovereign AI capabilities for millions of Arabic-speaking users.
