MineExplorer: Meituan LongCat evaluation reveals capability gaps in leading multimodal models in dynamic open worlds
Taking AI Out of the Greenhouse and Into a Dynamic World: MineExplorer Reveals an Overlooked Capability Gap in Leading Multimodal Models
AI BriefingThe Meituan LongCat team built MineExplorer, the first benchmark for minute-scale, long-horizon tasks in an open world. It includes 813 manually verified instances (1-hop to 4-hop) and a rule-based milestone framework for automated evaluation.
Source: WeChat Official Account: LongCat · mp.weixin.qq.com