Hala-9B: benchmark results
KAUST's open-weight Hala-9B (September 2025), an Arabic-centric instruction and translation model fine-tuned from QCRI's Fanar-1-9B-Instruct on the 4.6-million-example Hala-SFT dataset. Provider: KAUST. Released 2025-09-13. Access: Open.
Aggregate Rating 1579 ± 25, rank #517 of 1635 rated models, from 149 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Arabic LLM - Aratrust Privacy | 100 | Accuracy (%) | 99.7 |
| Open Arabic LLM - Aratrust Offensive | 97.1 | Accuracy (%) | 98.8 |
| Open Arabic LLM - Aratrust Unfairness | 96.36 | Accuracy (%) | 98.4 |
| Open Arabic LLM - Aratrust Average | 89.98 | Accuracy (%) | 97.8 |
| Open Arabic LLM - Alghafa Multiple Choice Facts Truefalse Balanced Task | 96 | Accuracy (%) | 97.2 |
| Open Arabic LLM - Alghafa Multiple Choice Rating Sentiment NO Neutral Task | 83.21 | Accuracy (%) | 95.7 |
| Open Arabic LLM - Alghafa Multiple Choice Rating Sentiment Task | 56.31 | Accuracy (%) | 93.8 |
| Open Arabic LLM - Arabic MMLU Arabic Language (Grammar) | 66.85 | Accuracy (%) | 93.8 |
| Open Arabic LLM - Alghafa Multiple Choice Grounded Statement Soqal Task | 94 | Accuracy (%) | 93.5 |
| Open Arabic LLM - Alghafa Average | 78.17 | Accuracy (%) | 93.2 |
| Open Arabic LLM - Arabic MMLU LAW (Professional) | 77.71 | Accuracy (%) | 92.3 |
| Open Arabic LLM - Madinah QA Arabic Language (Grammar) | 67.12 | Accuracy (%) | 91.7 |
Interactive version: theaggregate.ai/model?slug=hala-9b · How It Works · Data refreshed daily, snapshot 2026-11-05.