VitaBench 2.0, open-sourced by Meituan LongCat, is the first benchmark targeting
long-term dynamic user modeling in real-world scenarios. It systematically
evaluates large language models' ability to deliver personalization and
activity across extended, real-time user interactions.
