According to CoinMeta, Tencent Hunyuan, in collaboration with Tsinghua University and Peking University, launched iwc-bench to specifically evaluate the actual performance of web pages generated by AI. This benchmark includes 369 real user requirements and 5088 acceptance criteria, and 17 models generated a total of 6,273 web page applications. In 197 sets of manual reviews, iwc-bench matched human choices in 168 cases, achieving a consistency rate of 85.3%. The results show that a webpage "looking good" is not the same as being "truly user-friendly."