Mininglamp Launches $15,000 Global Challenge for AI Web Agents

The competition addresses a significant hurdle in artificial intelligence: the transition of Web Agents from controlled laboratory settings to functional, real-world business applications. Current benchmarks often rely on simplified simulations that fail to capture the volatility of the open web. To bridge this gap, Mininglamp Technology developed the WebRetriever benchmark, which covers 800 live websites and 1,550 cross-industry tasks. Research associated with the benchmark indicates that even leading models struggle with end-to-end task completion, often falling short of a 20% success rate in realistic scenarios.

Participants will focus on the End-to-End Task Protocol, which requires agents to navigate live pages, interact with elements, and extract multimodal data autonomously. The competition is open to individuals and teams of any nationality. The automated evaluation framework uses the NavEval system, which boasts a 91.2% agreement rate with human expert judgments. Registration remains open until August 7, 2026, with the final results scheduled for announcement in September. The initiative is supported by academic partners including Peking University and the Institute of Automation of the Chinese Academy of Sciences.