LitReview Arena: Evaluating Literature Review Agents with Battle-Style Peer Review Platform

LitReview Arena evaluates automated literature reviews through a peer-review platform where domain experts compare anonymized AI-generated drafts against human work. In evaluations involving 3,000 expert judgments, the strongest current systems won only 23.0% of decisive matches on overall utility, though agentic models like Sonar Deep Research outperformed base models by over 60%.