QIMMA - ArabLegalQA: leaderboard
ArabLegalQA open-ended questions on Saudi law, as run in the QIMMA Arabic leaderboard. Generated answers are scored by BERTScore F1 against references (0-100, higher is better); BERTScore has a high floor, so models span a narrow range.
Metric: BERTScore F1 (%). Source: huggingface.co. Saturation forecast: Not forecast. 83 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | granite-4.0-h-3B-ar | 74.63 | #1002 |
| 2 | Gemma 4 31B (IT) | 73.48 | #178 |
| 3 | Fanar-1-9B-Instruct | 73.08 | #635 |
| 4 | Gemma 4 26B A4B (IT) | 73.05 | #293 |
| 5 | gemma-4-E4B-it | 72.94 | #577 |
| 6 | Gemma 4 E2B | 72.39 | #1062 |
| 7 | Gemma 3 27B (IT) | 72.18 | #528 |
| 8 | Gemma 4 E4B | 72.07 | #762 |
| 9 | LFM2 8B A1B | 72 | #1248 |
| 10 | Gemma 3 4B (IT) | 71.88 | #997 |
| 11 | LFM2 24B A2B | 71.82 | #1362 |
| 12 | granite-4.0-h-micro | 71.68 | #1078 |
| 13 | ALLaM-7B-Instruct-preview | 71.5 | #724 |
| 14 | Yehia-7B-preview | 71.5 | #659 |
| 15 | AceGPT-v2-32B-Chat | 71.46 | #518 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=qimma-arablegalqa · How It Works · Data refreshed daily, snapshot 2026-11-05.