QIMMA - AraTrust: leaderboard
AraTrust multiple-choice questions on trustworthiness in Arabic, covering truthfulness, ethics, privacy, illegal activities, health, unfairness and offensive language, as run in QIMMA. Scored with LightEval log-likelihood over the answer options with character-length normalization rather than generated answers, so reasoning-tuned models can land near chance; accuracy 0-100, higher is better.
Metric: Accuracy (%, log-likelihood multiple choice). Source: huggingface.co. Saturation forecast: Not forecast. 83 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | AIC-1 | 90.61 | #235 |
| 2 | Jais-2-70B-Chat | 90.23 | #430 |
| 3 | Qwen 3.5 397B A17B | 90.04 | #154 |
| 4 | Gemma 4 31B | 90.04 | #212 |
| 5 | gemma-3-27B-pt | 89.08 | #505 |
| 6 | Hala-9B | 88.89 | #517 |
| 7 | Qwen 2.5 72B Instruct | 88.51 | #469 |
| 8 | Qwen 3.5 35B A3B Base | 88.51 | #319 |
| 9 | Fanar-1-9B-Instruct | 88.51 | #635 |
| 10 | Jais-2-8B-Chat | 87.55 | #802 |
| 11 | gemma-3-12B-pt | 87.36 | #686 |
| 12 | Yehia-7B-preview | 87.36 | #659 |
| 13 | Qwen 3.5 122B A10B | 86.97 | #184 |
| 14 | AceGPT-v2-32B-Chat | 86.97 | #518 |
| 15 | Qwen 3.5 27B | 86.59 | #192 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=qimma-aratrust · How It Works · Data refreshed daily, snapshot 2026-11-05.