ResearchBenchmarksAgentic Task Solving on AB-DBFollow51.1Pass@3A³35.18839.31943.4547.581May 8, 2026Evaluation ResultsMethodMethodLinksPass@3Pass@5A³variant=Vanillavariant=Vanilla2026.0551.155.3A³variant=σ-Revealvariant=σ-Reveal2026.0550.355GSPO2026.0542.845.8RetroAgent2026.0541.844.7ReACT2026.0538.241.2HGPO2026.0537.841.2LATS2026.0537.741.2rStar2026.0536.141.2GiGPO2026.0535.838.9