AIRS-Bench Asks Whether AI Agents Can Do Frontier ML Research

A new benchmark tests AI agents on real machine learning research tasks — not toy problems — aiming to measure whether autonomous systems can match human researchers.