Alibaba's research team has been awarded the Best Resource Paper at the ACL 2026 conference for their work on Deep Research Agents. The paper highlights significant flaws in current AI agents' ability to reason with complex real-world rules and introduces a new benchmark, HSCodeComp, to evaluate these capabilities. This benchmark, focused on accurately classifying goods with 10-digit HS Codes, revealed that even the best-performing agent systems achieved only about 45% accuracy, far below human expert levels. The research also identified structural bottlenecks in agent architecture, such as long reasoning chains and insufficient domain knowledge, as key limitations. AI
IMPACT Establishes a new standard for evaluating AI agents in complex rule-based tasks, potentially accelerating the development of more reliable AI systems for professional domains.
RANK_REASON Academic paper award at a top-tier NLP conference.
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
