HELM Arabic - Madinah Qa: leaderboard
The "Madinah Qa" subtask of HELM Arabic. Stanford HELM's Arabic project: Modern Standard Arabic knowledge, reasoning and understanding scenarios (AlGhafa, Arabic MMLU, exams and native-authored sets) run through the HELM harness with exact-match style metrics.
Metric: EM. Source: crfm.stanford.edu. Saturation forecast: Around May 2027. 41 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.7 | 90.95 | #50 |
| 2 | Gemini 2.5 Flash (Non-reasoning) | 90.22 | #242 (Gemini 2.5 Flash) |
| 3 | Claude Sonnet 4.6 | 89.94 | #92 |
| 4 | Gemma 4 31B (IT) | 89.27 | #178 |
| 5 | GPT-5.4 (2026-03-05) | 86.96 | #58 |
| 6 | GPT-4.1 (2025-04-14) | 86.58 | #250 |
| 7 | GPT-5.1 (2025-11-13) | 85.93 | #105 |
| 8 | Gemini 2.5 Flash Lite (Non-reasoning) | 85.84 | #421 (Gemini 2.5 Flash Lite) |
| 9 | Claude Haiku 4.5 (20251001) | 84.81 | #258 |
| 10 | Llama 4 Maverick Instruct FP8 | 83.49 | #412 |
| 11 | Mistral Large 3 | 82.75 | #404 |
| 12 | GPT-4.1 Mini | 82.6 | #351 |
| 13 | Qwen 3.5 397B A17B | 82.45 | #147 |
| 14 | GPT-5.4 Mini (2026-03-17) | 81.28 | #171 |
| 15 | Qwen 3 Next 80B A3B Instruct | 80.87 | #370 |
No result here: #4 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=helm-arabic-madinah-qa · How It Works · Data refreshed daily, snapshot 2026-10-23.