Arena-Hard

'Arena Hard' is a benchmark dataset used to evaluate the performance of models on challenging tasks that require a combination of translation specialization and general-purpose capabilities.

Papers using Arena-Hard (3)

Examining Reasoning LLMs-as-Judges in Non-Verifiable LLM Post-Training2026FuseChat-3.0: Preference Optimization Meets Heterogeneous Model Fusion2025Tower+: Bridging Generality and Translation Specialization in Multilingual LLMs2025