Open Arabic LLM - Arabic MMLU HT Formal Logic: leaderboard
Metric: Accuracy (%). Source: huggingface.co. 163 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen3-8B-Base | 61.9 |
| 2 | Qwen2.5-32B-Instruct-CFT | 59.52 |
| 3 | Gemma 3 27B (IT) | 58.73 |
| 4 | Llama 3.3 70B Instruct | 58.73 |
| 5 | lambda-qwen2.5-14B-dpo-test | 58.73 |
| 6 | Awqward2.5-32B-Instruct | 57.94 |
| 7 | tempmotacilla-cinerea-0308 | 57.94 |
| 8 | Qwen 2.5 32B Instruct | 57.14 |
| 9 | Qwen2.5-32B-Instruct-abliterated-v2 | 57.14 |
| 10 | Qwen 2.5 72B Instruct | 56.35 |
| 11 | Saka-14B | 56.35 |
| 12 | Qwentile2.5-32B-Instruct | 55.56 |
| 13 | Qwen 2.5 14B | 53.17 |
| 14 | recoilme-gemma-2-9B-v0.4 | 51.59 |
| 15 | Phi-4 Mini Instruct | 50.79 |
Interactive version: theaggregate.ai/benchmark?slug=open-arabic-llm-arabic-mmlu-ht-formal-logic · How It Works · Data refreshed daily, snapshot 2026-09-05.