AraGen Benchmark: 3C3H-based Arabic LLM Evaluation Leaderboard

AI Impact Summary

AraGen introduces a dynamic Arabic LLM evaluation framework built on the 3C3H measure, plus AraGen Benchmark and Leaderboard. It uses LLM-as-a-Judge to score model outputs across Correctness, Completeness, Conciseness, Helpfulness, Honesty, and Harmlessness, with three-month blind evaluation cycles to reduce data leakage. This sets a standard for comparing Arabic models across factuality and usability and creates a migration path for teams to integrate the 3C3H rubric into internal QA dashboards and external benchmarking. The approach, paired with private datasets before public release, signals a scalable model for multilingual benchmarking that could extend to other languages.