AraGen - Safety: leaderboard
AraGen 3C3H score on the 51-item Safety task of the v3 Arabic question set, averaging correctness, completeness, conciseness, helpfulness, honesty and harmlessness as judged by claude-sonnet-4. Shown x100, higher is better; runs that answered under 90% of the Safety items are excluded.
Metric: 3C3H Score (%). Source: huggingface.co. Saturation forecast: Around February 2027. 65 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.5 (20251101) | 91.09 | #81 |
| 2 | Claude Haiku 4.5 (20251001) | 89.71 | #264 |
| 3 | Llama 3.3 70B Instruct | 89.46 | #541 |
| 4 | Claude Opus 4 (20250514) | 88.32 | #164 |
| 5 | Claude 3.7 Sonnet (20250219) | 88.24 | #209 |
| 6 | Claude Sonnet 4.5 | 86.93 | #148 |
| 7 | Claude Sonnet 4 (20250514) | 86.36 | #223 |
| 8 | Mistral Small 3 | 85.87 | #630 |
| 9 | O1 (2024-12-17) | 83.25 | #157 |
| 10 | O4 Mini (2025-04-16) | 81.83 | #183 |
| 11 | Gemma 4 31B (IT) | 81.13 | #178 |
| 12 | Gemma 4 26B A4B (IT) | 81.05 | #293 |
| 13 | Jais-2-70B-Chat | 80.64 | #430 |
| 14 | O3 Mini (2025-01-31) | 80 | #249 |
| 15 | O1 Mini (2024-09-12) | 79.75 | #412 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=aragen-safety · How It Works · Data refreshed daily, snapshot 2026-11-05.