Transcribe Arabic

Cohere Transcribe Arabic: Frontier Speech Recognition for Arabic Speakers | Cohere North Mini Code. Cohere's first model for developers. Learn more

Jul 07, 2026

5 minute read

Meet Cohere Transcribe Arabic The world’s most accurate open-source model for Arabic speech recognition.

Key takeaways 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. Human reviewers preferred Cohere Transcribe Arabic to Whisper in 96% of tests. This is a proud advance in the region’s sovereign AI capabilities, bringing frontier performance to millions of Arabic-speakers.

Today, we are releasing Cohere Transcribe Arabic as an open-source model. Based on our 2B frontier Automatic Speech Recognition (ASR) model released earlier this year, Cohere Transcribe Arabic is built for the realities of Arabic in business and developer settings: dialect variation, bilingual Arabic-English speech, code-switching, and domain-specific vocabulary.

It is the most accurate, open-source Arabic speech-to-text model to date, outperforming leading alternatives, including Whisper and OmniASR, across dialects and common speech patterns. It also delivers substantial gains over Cohere Transcribe on both Arabic and bilingual Arabic-English audio.

Cohere Transcribe Arabic is available under the Apache 2.0 license . Developers can download the weights and read our quickstart implementations on Hugging Face, or access the hosted model through the Cohere API or Model Vault . One model, many voices Arabic is a remarkably rich language in both script and speech. More than 300 million people speak Arabic as their mother tongue, across roughly 30 recognized varieties shaped by distinct cultural, regional, and historical contexts. Saudi Arabia alone is home to three major dialect groups and many more linguistic subgroups.

This diversity, however, heavily complicates efforts towards normalisation. While Modern Standard Arabic (MSA) provides a de facto common written standard, everyday speech varies significantly across dialects. Morphological differences, regional pronunciation, and code-switching — the use of Arabic and non-Arabic vocabulary in the same conversation, often in professional settings — make a single, uniform approach to communication difficult. Image 1: a map of the Arabic-speaking world, highlighting some of the largest dialect families, their number of native speakers, and geographic distribution. Source: Ethnologue The challenge is especially stark in the development of natural language technology, such as ASR. How do you train a model that preserves dialectal nuance while remaining useful beyond a particular market? The result has been a frontier-language gap: Arabic remains under-served by state-of-the-art AI systems while English continues to dominate model development and evaluation.

To help narrow that gap, Cohere embarked on a simple mission: build an enterprise-ready solution that lets Arabic users speak in their natural voice.

We started with Cohere Transcribe (launched in March with leading English-language accuracy and broad multilingual coverage) and trained it extensively on data spanning Arabic dialects, professional language, code-switching, and varied acoustic conditions.

The result is a new state-of-the-art solution in how Arabic speech is captured, ready for production use and openly available to all. Unrivalled accuracy Cohere Transcribe Arabic achieves the lowest average word error rate (WER) of any open-source model on the Hugging Face Arabic ASR Leaderboard , with a WER of 25.87. This is a 2.45-point improvement over the previous leader, Meta’s OmniASR-LLM-7B, and an 11-point improvement over OpenAI’s Whisper Large V3.

Dataset Cohere Transcribe Arabic OmniASR 7B-LLM OpenAI Whisper Large V3 Cohere Transcribe

Average WER 25.87 28.32 36.86 30.67

SADA 37.47 41.61 55.96 60.11

Common Voice 5.82 9.75 17.83 8.17

MASC (clean) 15.54 19.69 24.66 8.66

MASC (noisy) 27.07 29.29 34.63 19.01

MGB-2 15.54 14.13 16.26 25.33

Casablanca 49.71 56.46 71.81 62.71

Image 2: the Open Universal Arabic ASR leaderboard as of 07.07.26. This public benchmark evaluates zero-shot multi-dialect generalization across six test sets spanning MSA, Egyptian, Gulf, Levantine, and Maghrebi dialects. See the latest leaderboard and details on the benchmark methodology.

The gains are broad-based. Cohere Transcribe Arabic delivers the best overall WER and ranks first on four of the six composite task sets. On Casablanca, for example, which evaluates conversational Arabic across eight dialects, it improves on OmniASR by nearly six points. On Common Voice, a crowd-sourced dataset covering 25 dialects, it reduces WER by more than two points from a low previous base.

The performance carried through to human evaluations with native Arabic speakers. Evaluators assessed transcription quality across three dimensions: Overall accuracy : How well the transcript captured both the form and semantic meaning of the audio Dialect faithfulness : Whether the model preserved the speaker’s dialect rather than converging toward Modern Standard Arabic Robustness to code-switching : Whether English terms were correctly transcribed in Latin script

Cohere Transcribe Arabic scored highest on all three dimensions compared with Whisper and Cohere Transcribe. In head-to-head evaluations, it was preferred over Whisper in 95.8% of tests. Image 3: human preference evaluations of Arabic transcripts, according to overall quality, faithfulness to the dialect used, and ability to handle code-switching elements. Scores were rated on a scale of 1-5, with 5 marking the highest quality. The model also improved significantly on Cohere Transcribe for English spoken with an Arabic accent, covering many workplace and second-language English use cases. Human evaluators preferred Cohere Transcribe Arabic over Cohere Transcribe in 77.2% of tests, and found it broadly comparable to Whisper on these inputs, with Cohere Transcribe Arabic preferred in 52.6% of tests. Image 4: human preference scores of Arabic transcripts in pairwise comparisons. A score of 50% or higher indicates that outputs from Cohere Transcribe Arabic were considered more accurate on average. Enterprise-ready Cohere Transcribe Arabic is built for high-throughput serving in production environments, where performance...