Cohere has introduced Cohere Transcribe Arabic, an open-source ASR model designed for Arabic speech recognition. The release focuses on speech-to-text problems that are especially common in Arabic, including dialect variety, bilingual Arabic-English conversations, code-switching, and specialized vocabulary.
The model has 2-billion-parameter and is being positioned by Cohere as the most accurate open-source Arabic speech-to-text system available. Cohere says it performs better than Whisper Large V3, the standard Cohere Transcribe model, and other systems in benchmarks.
Why Arabic Speech Recognition Needs A Focused Model
Arabic transcription is not one simple task. The source highlights several challenges that make Arabic speech recognition difficult for general-purpose ASR systems: many dialects, conversations that move between Arabic and English, code-switching, and domain-specific terminology.
Dialect variety matters because Arabic speech can differ widely depending on the speaker and context. A model built for Arabic transcription must be able to process more than one narrow form of spoken Arabic if it is going to be useful across real conversations.
Bilingual Arabic-English conversations add another layer of complexity. In many settings, speakers may use both languages in the same exchange. The source identifies this as a target problem for Cohere Transcribe Arabic, alongside code-switching, where language use can shift within a conversation.
Specialized vocabulary is also part of the model’s stated focus. That matters for speech-to-text work because names, technical terms, professional language, and field-specific phrasing can be harder to capture correctly than common everyday words.
What Cohere Released
Cohere Transcribe Arabic is a 2-billion-parameter ASR model. ASR stands for automatic speech recognition, the category of systems that convert spoken audio into written text.
The model is open-source and ships under the Apache 2.0 license. The source says it is available on Hugging Face and through the Cohere API, giving users more than one route to access it.
The release is framed around Arabic speech-to-text rather than general transcription alone. Based on the source, the model is aimed at the specific recognition problems that appear when Arabic speakers use different dialects, mix Arabic and English, or rely on specialized vocabulary.
- Model: Cohere Transcribe Arabic
- Type: open-source ASR model
- Scale: 2-billion-parameter
- License: Apache 2.0 license
- Access: Hugging Face and Cohere API
How Cohere Says It Performs
Cohere says Cohere Transcribe Arabic is the most accurate open-source Arabic speech-to-text system available. The company also says the model outscores Whisper Large V3, the standard Cohere Transcribe model, and other systems in benchmarks.
The source does not include the benchmark tables or scores, but it notes that more benchmarks and examples are available on the Cohere blog. That means the central performance claim in the article is clear, while the supporting details are pointed to separately.
The comparison with Whisper Large V3 is notable because the source presents it as one of the systems Cohere says its Arabic model beats. The comparison with the standard Cohere Transcribe model also suggests Cohere is distinguishing this release from its broader transcription system by emphasizing Arabic-specific performance.
What This Means For Arabic Transcription
The release adds a dedicated open-source option for Arabic speech recognition. For anyone evaluating Arabic speech-to-text tools, the key claim is not only that Cohere Transcribe Arabic handles Arabic, but that it was built around Arabic’s hardest transcription cases as described by Cohere.
Those cases include dialect variety, bilingual Arabic-English conversations, code-switching, and specialized vocabulary. Each one can make transcription more demanding because the model has to keep up with shifts in how people speak, which language they use, and what vocabulary appears in the audio.
The Apache 2.0 license and availability on Hugging Face make the model accessible as an open-source release. Access through the Cohere API gives another path for using the model without relying only on the Hugging Face route.
For now, the source’s main takeaway is straightforward: Cohere has released an Arabic-focused ASR model, says it leads open-source Arabic speech-to-text accuracy, and is making it available through both open-source distribution and its API.