HELM Arabic Enterprise - Arabic Legal Rag: leaderboard
The "Arabic Legal Rag" subtask of HELM Arabic Enterprise. Stanford HELM's Arabic Enterprise project: business-facing Arabic tasks — content generation, summarisation, translation, dialect handling and customer-support style requests — scored by rubric and judge metrics rather than exact match.
Metric: Score. Source: crfm.stanford.edu. Saturation forecast: Around November 2026. 35 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.6 | 99 | #92 |
| 2 | GPT-5.1 (2025-11-13) | 99 | #105 |
| 3 | Claude Opus 4.7 | 98 | #50 |
| 4 | Claude Haiku 4.5 (20251001) | 98 | #258 |
| 5 | DeepSeek V3.1 | 98 | #268 |
| 6 | Gemma 4 31B (IT) | 97.5 | #178 |
| 7 | GPT-5.4 Mini (2026-03-17) | 97.5 | #171 |
| 8 | GPT-5.4 (2026-03-05) | 97 | #58 |
| 9 | DeepSeek V4 Pro | 96.5 | #106 |
| 10 | Qwen 3.5 397B A17B | 96.5 | #147 |
| 11 | Mistral Large 3 | 96.5 | #404 |
| 12 | Mistral Large 2 (Nov) Instruct (2411) | 96.5 | #468 |
| 13 | GPT-5.4 Nano (2026-03-17) | 96.5 | #266 |
| 14 | Gemini 2.5 Flash (Non-reasoning) | 96 | #242 (Gemini 2.5 Flash) |
| 15 | Qwen 3.5 9B | 95 | #384 |
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-enterprise-arabic-legal-rag · How It Works · Data refreshed daily, snapshot 2026-10-23.