HELM Arabic - Aratrust - Privacy: leaderboard
HELM Arabic (release v2.2.0) AraTrust Privacy category: human-written Arabic three-option multiple-choice questions probing trustworthiness on privacy; 57 questions, exact match. Percent correct, 0-100, higher is better; guessing scores 33.3.
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 | 100 | #51 |
| 2 | Gemma 4 31B (IT) | 100 | #178 |
| 3 | Llama 4 Maverick Instruct FP8 | 100 | #424 |
| 4 | GPT-5.4 (2026-03-05) | 100 | #63 |
| 5 | Claude Sonnet 4.6 | 98.25 | #94 |
| 6 | GPT-4.1 Mini | 98.25 | #363 |
| 7 | Qwen 3.5 9B | 98.25 | #391 |
| 8 | Qwen 3.5 397B A17B | 98.25 | #154 |
| 9 | GPT-4.1 (2025-04-14) | 98.25 | #257 |
| 10 | GPT-5.1 (2025-11-13) | 98.25 | #107 |
| 11 | Gemini 2.5 Flash Lite (Non-reasoning) | 98.25 | #432 (Gemini 2.5 Flash Lite) |
| 12 | AceGPT-v2-32B-Chat | 98.25 | #518 |
| 13 | GPT-5.4 Mini (2026-03-17) | 98.25 | #180 |
| 14 | Llama 4 Scout Instruct | 96.49 | #547 |
| 15 | Claude Haiku 4.5 (20251001) | 96.49 | #264 |
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-aratrust-privacy · How It Works · Data refreshed daily, snapshot 2026-11-05.