Open Universal Arabic ASR Leaderboard: leaderboard
Multi-dialect Arabic ASR leaderboard reporting average WER/CER and dataset-level WER/CER on SADA, Common Voice, MASC, MGB-2, and Casablanca.
Metric: Average WER (lower is better). Source: huggingface.co. Saturation forecast: Around May 2027. 37 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen3 Omni 30B A3B Instruct | 30.71 | #383 |
| 2 | gemma-4-E4B-it | 32.98 | #566 |
| 3 | Voxtral-Small-24B-2507 | 34.47 | #581 |
| 4 | Gemma 4 E2B (IT) | 35.87 | #781 |
| 5 | Whisper Large-v3 | 36.86 | #1271 |
| 6 | Voxtral-Mini-3B-2507 | 42.58 | #1020 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=open-universal-arabic-asr-leaderboard · How It Works · Data refreshed daily, snapshot 2026-10-20.