NEWS · MODELS · #416
Cohere releases open-source "Transcribe Arabic" ASR model claiming best open-weight Arabic accuracy
Cohere published Transcribe Arabic, an open-source Arabic automatic speech recognition (ASR) model (Apache 2.0) based on its 2B ASR work. Cohere says the model achieves a 25.87 WER on the Hugging Face Arabic ASR leaderboard, outperforming Meta’s OmniASR-LLM-7B and OpenAI’s Whisper Large V3, and was preferred to Whisper in about 96% of human evaluation tests; weights are available on Hugging Face and via the Cohere API/Model Vault.
KEY POINTS
- Cohere published Transcribe Arabic, an open-source Arabic automatic speech recognition (ASR) model (Apache 2.0) based on its 2B ASR work.
- Cohere says the model achieves a 25.87 WER on the Hugging Face Arabic ASR leaderboard, outperforming Meta’s OmniASR-LLM-7B and OpenAI’s Whisper Large V3, and was preferred to Whisper in about 96% of human evaluation tests; weights are available on Hugging Face and via the Cohere API/Model Vault.
- An open-source, enterprise-ready Arabic ASR model with reported leaderboard-leading WER and strong human-eval preference can materially improve Arabic speech tooling and reduce dependence on proprietary systems.
DECISION BRIEF
CONFIDENCE
WHAT CHANGED
Cohere published Transcribe Arabic, an open-source Arabic automatic speech recognition (ASR) model released under the Apache 2.0 license. Cohere states the model is based on its 2B ASR model, that it outperforms leading alternatives (including Whisper and OmniASR) for Arabic transcription, that human reviewers preferred it to Whisper in ~96% of tests, and that model weights are available on Hugging Face and via Cohere’s API/Model Vault (Cohere sources).
WHY NOW
Cohere is making an enterprise-oriented, open-source Arabic ASR model available for download and hosted inference while claiming leaderboard-leading accuracy and strong human-eval preference. Those two elements (open-weights availability + provider claims of superior accuracy/preference) are the proximate reasons this release could affect Arabic ASR tooling and choices now.
WHO IS AFFECTED
Directly affected parties called out in the sources: developers (weights and quickstarts available on Hugging Face), enterprises/teams building speech applications (Cohere frames the model as fit for enterprise use), and Arabic-speaking users/markets (Cohere references dialectal coverage and the region’s speakers). The Hugging Face Open ASR leaderboard community is also implicated by leaderboard claims.
CONFIRMED
All explicit, source-backed facts from the provided excerpts: - Cohere released "Cohere Transcribe Arabic" as an open-source Arabic ASR model under the Apache 2.0 license. - Cohere says the model is based on its 2B ASR model. - Cohere states the model "achieves the highest accuracy for Arabic-language transcription among open-weights models" and that it outperforms alternatives including Whisper and OmniASR across dialects and common speech patterns. - Cohere reports human reviewers preferred Cohere Transcribe Arabic to Whisper in 96% of tests. - Cohere provides model weights on Hugging Face and access via Cohere API / Model Vault. - Separately, Cohere’s Transcribe (multi-language) is described as ranking #1 on the Hugging Face Open ASR Leaderboard with an average WER of 5.42%.
UNCERTAIN
Key uncertainties or missing/conflicting evidence in the supplied excerpts: - No third-party or independent benchmark numbers for Transcribe Arabic are provided in the excerpts; claims of outperforming alternatives and the 96% human-preference result are presented by Cohere without external validation shown in the excerpts. - The specific numeric Arabic WER cited in other summaries (e.g., 25.87) is not present in the supplied excerpts and therefore not confirmed here (missing evidence). - Details on datasets, test conditions, or evaluation methodology for the Arabic comparisons (which dialects, dataset splits, normalization rules) are not provided in the excerpts (missing evidence). - Operational performance (inference latency/cost on specific hardware for Arabic workloads) and privacy/data provenance details are not specified in the excerpts (missing evidence).
WHAT TO WATCH
Observable signals to monitor next (concrete, verifiable): - Hugging Face Open ASR Leaderboard updates for Arabic-specific entries and numeric WERs (verify whether Transcribe Arabic appears and reported WER) (relevant to source 523/479). - Publication or release of evaluation details from Cohere (datasets, test methodology, examples) that substantiate the 96% human-preference claim (source 479 currently asserts the result). - Independent third-party benchmark reports or academic/industry evaluations comparing Transcribe Arabic, Whisper, and OmniASR on standard Arabic datasets. - Repo activity and download/usage metrics on Hugging Face for the Transcribe Arabic model and issues/PRs that surface real-world limitations or bugs (weights available per source 479). - Announcements or documentation from Cohere about production/inference performance or case studies showing enterprise deployments (source 479 positions the model as enterprise-ready).
WHY IT MATTERS
An open-source, enterprise-ready Arabic ASR model with reported leaderboard-leading WER and strong human-eval preference can materially improve Arabic speech tooling and reduce dependence on proprietary systems.
EVIDENCE MAP
4Editorial claims linked to specific sources, with support, contradiction and context shown separately.
Cohere released Transcribe Arabic as an open-source Arabic ASR model under the Apache 2.0 license.
SUPPORTEDChecked 2026-09-25 · 1 supporting
Cohere says Transcribe Arabic is the most accurate open-weights model for Arabic speech recognition and outperforms Whisper and OmniASR on Arabic transcription.
SUPPORTEDChecked 2026-09-25 · 1 supporting
Cohere reports that human reviewers preferred Transcribe Arabic to Whisper in about 96% of tests.
SUPPORTEDChecked 2026-09-25 · 1 supporting
Developers can download Transcribe Arabic weights from Hugging Face or access the model via the Cohere API or Model Vault.
SUPPORTEDChecked 2026-09-25 · 1 supporting
SOURCES & TIMELINE
1Unmatched transcription accuracy — our first step toward powering enterprise speech intelligence in North. Cohere is announcing Transcribe, a state-of-the-art automatic speech recognition (ASR) model that is open source and available today for download . Speech is rapidly becoming a core modality for AI-enabled workloads and automations — from meeting transcription and speech analytics to real-time customer support a…
Our fastest and most powerful language model yet. Command A+ is an open-source enterprise workhorse built for complex reasoning, multimodal and multilingual agentic tasks — all while running on as little as two H100 GPUs. Today, we’re releasing Command A+ open-source. A mixture-of-experts (MoE) model, Command A+ is an efficient, versatile, and privately deployable LLM built for high-performance agentic tasks with min…
The world’s most accurate open-source model for Arabic speech recognition. 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 …