HELM Arabic - Alghafa - Meta AR Dialects: leaderboard
HELM Arabic (release v2.2.0) AlGhafa Meta AR Dialects task: four-option Belebele reading-comprehension questions in Arabic dialects, from the native AlGhafa Arabic multiple-choice benchmark; 1,000 items, exact match. Percent correct, 0-100, higher is better.
Metric: Exact Match (%). Source: crfm.stanford.edu. Saturation forecast: Estimated already saturated. 41 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.7 | 91.6 | #51 |
| 2 | Gemma 4 31B (IT) | 90.8 | #178 |
| 3 | Gemini 2.5 Flash (Non-reasoning) | 90.1 | #255 (Gemini 2.5 Flash) |
| 4 | GPT-5.4 (2026-03-05) | 89.8 | #63 |
| 5 | Claude Sonnet 4.6 | 88.9 | #94 |
| 6 | GPT-5.1 (2025-11-13) | 88.6 | #107 |
| 7 | GPT-4.1 (2025-04-14) | 87.6 | #257 |
| 8 | Qwen 3.5 397B A17B | 87.5 | #154 |
| 9 | Gemini 2.5 Flash Lite (Non-reasoning) | 86.6 | #432 (Gemini 2.5 Flash Lite) |
| 10 | Llama 4 Maverick Instruct FP8 | 85.7 | #424 |
| 11 | Claude Haiku 4.5 (20251001) | 84.8 | #264 |
| 12 | GPT-5.4 Mini (2026-03-17) | 84.6 | #180 |
| 13 | Mistral Large 3 | 84.4 | #410 |
| 14 | Qwen 2.5 72B Instruct | 83.8 | #469 |
| 15 | GPT-4.1 Mini | 83.6 | #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-alghafa-meta-ar-dialects · How It Works · Data refreshed daily, snapshot 2026-11-05.