HELM Arabic - Arabic Exams - Social: leaderboard
HELM Arabic (release v2.2.0) Arabic EXAMS, Social questions: Arabic high-school examination multiple-choice questions from the Arabic subset of EXAMS (OALL mirror of the AceGPT version); 272 questions, exact match. Percent correct, 0-100, higher is better.
Metric: Exact Match (%). Source: crfm.stanford.edu. Saturation forecast: Around 2031. 41 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Llama 3.3 70B Instruct | 55.51 | #541 |
| 2 | Qwen 2.5 72B Instruct | 51.1 | #469 |
| 3 | Claude Haiku 4.5 (20251001) | 50.74 | #264 |
| 4 | Gemini 2.5 Flash (Non-reasoning) | 50.74 | #255 (Gemini 2.5 Flash) |
| 5 | Llama 4 Maverick Instruct FP8 | 50.37 | #424 |
| 6 | GPT-5.1 (2025-11-13) | 50.37 | #107 |
| 7 | Claude Opus 4.7 | 50 | #51 |
| 8 | Llama 4 Scout Instruct | 50 | #547 |
| 9 | GPT-4.1 (2025-04-14) | 50 | #257 |
| 10 | Mistral Large 3 | 49.63 | #410 |
| 11 | Gemini 2.5 Flash Lite (Non-reasoning) | 49.63 | #432 (Gemini 2.5 Flash Lite) |
| 12 | GPT-5.4 (2026-03-05) | 49.26 | #63 |
| 13 | DeepSeek V3.1 | 48.9 | #275 |
| 14 | Claude Sonnet 4.6 | 48.53 | #94 |
| 15 | GPT-4.1 Mini | 47.79 | #363 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=helm-arabic-arabic-exams-social · How It Works · Data refreshed daily, snapshot 2026-11-05.