WebArena

WebArena is a benchmark dataset that contains traces from large language model (LLM) agents executing tool-using tasks, used to evaluate the effectiveness of online warning monitors in detecting potential failures during these tasks.

Papers using WebArena (7)

WebChoreArena: Evaluating Web Browsing Agents on Realistic Tedious Web Tasks2025 · 2 citesLearn-by-interact: A Data-Centric Framework for Self-Adaptive Agents in Realistic Environments2025 · 2 citesPrefixGuard: From LLM-Agent Traces to Online Failure-Warning Monitors2026Structured Distillation of Web Agent Capabilities Enables Generalization2026FocusAgent: Simple Yet Effective Ways of Trimming the Large Context of Web Agents2025R-WoM: Retrieval-augmented World Model For Computer-use Agents2025Auto-Intent: Automated Intent Discovery and Self-Exploration for Large Language Model Web Agents2024