This work investigates whether MLLMs can be used as unified evaluator for AI-generated videos and proposes a benchmark to assess this ability.