AIRA₂ Research Agent Tops MLE-bench with 81.5% Win Rate Above Human Baseline

Martin Josifoski's team released AIRA₂, an AI research agent that achieved state-of-the-art 81.5% win rate on MLE-bench-30, beating the 72.7% human baseline. The authors also clarified via ablations that its validation-set usage (model selection after HPO) has limited impact.

2026-09-25 ~ 2026-09-25 · 2 related posts

• AIRA₂ Research Agents Hit 81.5% on MLE-bench-30, Beating Prior SoTA of 72.7% — mariofilhoml · 2026-09-25
• Overfitting the Validation Set Is Rarely a Big Issue, AIRA₂ Ablations Show — mariofilhoml · 2026-09-25