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 proactivity across extended, real-time user interactions.

2026-06-25

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 proactivity across extended, real-time user interactions.