AraGen — leaderboard
Arabic generative benchmark using the 3C3H metric (Correctness, Completeness, Conciseness, Helpfulness, Honesty, Harmlessness) to evaluate Arabic LLM capabilities across QA, grammar, reasoning, and safety tasks.
Metric: 3C3H Score (%). Source: huggingface.co. Status: saturation imminent. 69 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | O1 (2024-12-17) | 84.29 |
| 2 | GPT-5 | 84.25 |
| 3 | O3 (2025-04-16) | 82.19 |
| 4 | Claude Opus 4 (20250514) | 80.96 |
| 5 | Claude Opus 4.5 (20251101) | 80.29 |
| 6 | Claude Sonnet 4.5 | 78.17 |
| 7 | Claude 3.7 Sonnet (20250219) | 78.16 |
| 8 | Claude Sonnet 4 (20250514) | 75.58 |
| 9 | GPT-4.1 | 74.54 |
| 10 | Gemma 4 31B (IT) | 72.71 |
| 11 | O4 Mini (2025-04-16) | 70.6 |
| 12 | O1 Mini (2024-09-12) | 70.46 |
| 13 | Gemini 2.5 Pro (Preview 05-06) | 68.37 |
| 14 | Gemma 4 26B A4B (IT) | 67.35 |
| 15 | Claude Haiku 4.5 (20251001) | 65.8 |
Interactive version: theaggregate.ai/benchmark?slug=aragen · How the rankings work · Data refreshed daily, snapshot 2026-07-22.