Arabic ASR WER: Open Universal Arabic ASR Leaderboard (average WER)
Average word error rate across the six multi-dialect Arabic ASR test sets used by the Open Universal Arabic ASR Leaderboard: SADA, Common Voice 18.0, MASC clean-test, MASC noisy-test, MGB-2, and Casablanca. Lower is better.
Models ranked
1
tracked on this benchmark
Score band
25.9 – 25.9
best → lowest tracked
Snapshot trend
—
need ≥2 snapshots
Leaderboard
Tracked models ranked by Average WER (%) (lower is better).
Notes: Variant: CohereLabs/cohere-transcribe-arabic-07-2026. Evaluator/leaderboard owner: Elm Research Center (Elm Company). Harness: Open Universal Arabic ASR Leaderboard; public Gradio configuration and app.py state that it reports WER/CER and ranks by average WER across SADA, Common Voice 18.0, MASC clean-test, MASC noisy-test, MGB-2, and Casablanca. The public leaderboard config showed 25.87 Average WER for this exact model on access date. Recommended seed value: 25.87. Harness source: https://huggingface.co/spaces/elmresearchcenter/open_universal_arabic_asr_leaderboard/raw/main/app.py
How to read this benchmark
This benchmark scores models where lower is better. Use scores for directional filtering and shortlisting, not universal quality ranking; then validate pricing, context window, provider availability, and fit for your workload.
Trust this score when
- There is a fresh timestamped snapshot (or multiple snapshots) for this benchmark.
- The model list covers the same version family you can actually deploy today.
- Top candidates overlap with your required routing and feature requirements.
Be cautious when
- There is only one benchmark snapshot or the dataset appears stale.
- The benchmark metric direction is opposite of your decision objective.
- The score difference between options is narrow and likely within implementation variance.
Related benchmarks
Last reviewed: Jul 13, 2026