HELM Arabic - Aratrust - Trustfulness: leaderboard
Metric: Exact Match (%). Source: crfm.stanford.edu. 41 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-4.1 (2025-04-14) | 91.03 |
| 2 | Claude Opus 4.7 | 91.03 |
| 3 | Gemini 2.5 Flash (Thinking) | 91.03 |
| 4 | GPT-5.4 (2026-03-05) | 91.03 |
| 5 | Qwen 3 Next 80B A3B Instruct | 89.74 |
| 6 | DeepSeek V3.1 | 88.46 |
| 7 | Mistral Large 3 | 88.46 |
| 8 | Claude Sonnet 4.6 | 85.9 |
| 9 | Llama 4 Scout Instruct | 84.62 |
| 10 | Qwen 3.5 397B A17B | 84.62 |
| 11 | Gemma 4 31B (IT) | 83.33 |
| 12 | Llama 4 Maverick Instruct FP8 | 83.33 |
| 13 | Gemini 2.5 Flash Lite (Thinking) | 83.33 |
| 14 | Qwen 2.5 72B Instruct | 82.05 |
| 15 | Claude Haiku 4.5 (20251001) | 82.05 |
Interactive version: theaggregate.ai/benchmark?slug=helm-arabic-aratrust-trustfulness · How It Works · Data refreshed daily, snapshot 2026-09-19.