WuYuEval: A new benchmark tests LLMs in waste management

A team of researchers from Tsinghua University and collaborators has introduced **WuYuEval**, a multi-level benchmark designed to evaluate the capabilities of large language models (LLMs) specifically in solid waste management (SWM). Unlike general-purpose benchmarks, WuYuEval bridges the gap between broad language understanding and the specialized, constraint-driven decision-making required in SWM. The benchmark is divided into two modules: a Foundation Module with 4,590 closed-ended multiple-choice questions across six task types and eight domain categories, and an Expert Module with 247 scenario-based open-ended questions that involve multi-objective optimization, constraint trade-offs, and system design.

Testing 33 LLMs, the researchers found significant variability in performance. The top-performing model achieved 94.64% accuracy on the Foundation Module, but the average accuracy dropped from 84.14% on easy questions to just 42.50% on hard ones. The lowest performance areas included calculation tasks, experimental design, urban planning, and open-ended expert tasks. The study also found that reasoning-oriented modes improved performance only when anchored to explicit constraints; otherwise, they risked drifting from decisive answers. This underscores the need for professional reasoning chains and constraint control in SWM-oriented AI development.