Command Palette
Search for a command to run...

Cohere Launches Open Source Arabic Speech Model That Beats Whisper In 96% Of Human Tests

aiai-modelingai-model-releasesai-open-modelsai-research-evals 7 posts · 5 accounts

Cohere launched Transcribe Arabic, a new open source speech recognition model licensed under Apache 2.0 that the company says tops the Open Universal Arabic ASR Leaderboard. The model claims superior transcription accuracy compared to existing open-weight systems, beating Whisper in 96% of human preference tests.

Designed to handle code-switching, multiple dialects, and Arabic-accented English, the architecture targets developers and businesses serving millions of Arabic speakers. The weights are available for download, allowing for direct integration into voice transcription and localization applications.

From the sources (7 posts)

@cohere

We’ve built Cohere Transcribe Arabic, the world’s most accurate open-source model for Arabic speech recognition. Available under Apache 2.0

@huggingface

RT @cohere: We’ve built Cohere Transcribe Arabic, the world’s most accurate open-source model for Arabic speech recognition. Available und…

@clementdelangue

RT @cohere: We’ve built Cohere Transcribe Arabic, the world’s most accurate open-source model for Arabic speech recognition. Available und…

@aidangomez

RT @cohere: We’ve built Cohere Transcribe Arabic, the world’s most accurate open-source model for Arabic speech recognition. Available und…

@jayalammar

نطلق اليوم نموذج @cohere Transcribe العربي، أفضل نموذج مفتوح المصدر لتحويل الكلام العربي من الصوت إلى نص، مع مراعاة اللهجات العربية المختلفة. يتصدر معيار Open Universal Arabic ASR Leaderboard. صُمم النموذج لدعم حالات تحويل الكلام بين العر

@cohere

Our model tops the Open Universal Arabic ASR Leaderboard, outperforming Whisper and OmniASR in transcription accuracy. Human reviewers also preferred us to Whisper in 96% of tests.

@cohere

Transcribe Arabic is built to bring frontier capabilities to the millions of Arabic speakers in business and developer communities. Built to handle code-switching, multiple dialects, and Arabic-accented English, we’re taking real-life trans

Preview built on a synthetic news corpus (16 weeks, Apr–Jul 2026). Impact calls are model reads, not price data.

About Archive