WebThinker

RUC-NLPIR's deep-research agent for large reasoning models (NeurIPS 2025; ~300 citations, 1.5K★): a Deep Web Explorer and an autonomous think-search-draft workflow with RL-based trajectory refinement. The wave it launched continued with ARPO agent-RL (ICLR 2026, joint with Kuaishou), DeepAgent (WWW 2026 Oral), and the Arbor autonomous research agent (June 2026).