ResearchBenchmarksAgentic Task Solving on DataBenchFollow92.7Pass@3A³70.02875.91481.887.686May 8, 2026Evaluation ResultsMethodMethodLinksPass@3Pass@5A³variant=σ-Revealvariant=σ-Reveal2026.0592.796GSPO2026.0589.994.7A³variant=Vanillavariant=Vanilla2026.0588.493.3RetroAgent2026.0585.789.3ReACT2026.0579.782.7LATS2026.0577.685.3GiGPO2026.0577.284rStar2026.0572.580HGPO2026.0570.981.3