Cohere Releases Open-Source Arabic ASR Model, Topping Leaderboards

On July 7, Cohere officially launched Transcribe Arabic, an Arabic automatic speech recognition (ASR) model focusing on high accuracy and open-source accessibility. Released under the Apache 2.0 license, the model targets Arabic business and developer communities to enhance transcription capabilities in real-world scenarios.

Key Details

The specific model version is cohere-transcribe-arabic-07-2026, now available on HuggingFace and featured on the trending list. Beyond supporting various Arabic dialects, the model handles code-switching (mixing Arabic and English) as well as Arabic-accented English. Furthermore, it natively runs on the mlx-audio framework, supporting Python and other environments.

Performance and Reactions

Cohere claims the model has topped the Open Universal Arabic ASR Leaderboard, surpassing mainstream models like Whisper and OmniASR in transcription accuracy. In human review tests, reviewers preferred Cohere's transcription over Whisper in 96% of cases. AI researcher JayAlammar also highlighted the model's ability to handle dialects and transcribe speech between Arabic and English.

2026-07-07 ~ 2026-07-08 · 8 related posts

• [source] Cohere Releases Open-Source Arabic Speech Recognition Model — cohere · 2026-07-07
• Cohere Model Tops Arabic ASR Leaderboard — cohere · 2026-07-07
• Cohere Launches Arabic Transcription Product — cohere · 2026-07-07
• Cohere Releases Open-Source Arabic Speech Recognition Model — JayAlammar · 2026-07-07
• [source] Cohere Releases Arabic Speech-to-Text Model — Recoil42 · 2026-07-08
• Cohere Launches Transcribe Arabic Model — itsSandraKublik · 2026-07-08
• [source] Cohere Speech-to-Text Model Hits Trending — CohereLabs · 2026-07-08

1 near-duplicate retellings: JayAlammar