Woofun AI reports that Meituan's LongCat team has released VitaBench 2.0 under an open-source license, succeeding the version launched in October last year. The benchmark is designed to evaluate agents in real-life scenarios by modeling long-term, dynamic user behavior. It systematically assesses large language models' capabilities for personalization and proactivity during sustained, real-world interactions.