
ResearchArena: Are AI Agents Ready for Scientific Discovery?
ResearchArena evaluates off-the-shelf AI agents on their ability to conduct full research loops, from ideation to paper writing. The study reveals that while agents can produce plausible-looking manuscripts, they struggle significantly with experimental rigor and factual accuracy.





