Tencent Hunyuan launches E-Bench, a benchmark for evaluating intelligent agents.

Tencent Hunyuan, in collaboration with Tsinghua AIR and Southeast University, launched the E-Bench intelligent agent evaluation benchmark. Based on games like Honor of Kings, QQ Music, and Tencent Meeting, it constructs a fully synthetic virtual environment with 323 state-modification tasks and over 76,000 data rows. Through information gap and tool gap design, it prevents models from taking shortcuts based on memory. E-Bench evaluated 11 cutting-edge models with an average success rate of only 54.56%, with the strongest model, Avg@3, achieving 73.79%, and Pass³ below 60%.