ResearchBenchmarksSTRING on DataBenchFollow77.9LLM Judge AccuracyA343.5852.4961.470.31May 8, 2026Evaluation ResultsMethodMethodLinksLLM Judge AccuracyA3Harness Context=σ-RevealHarness Context=σ-Reveal2026.0577.9A3Harness Context=VanillaHarness Context=Vanilla2026.0574.6GSPO2026.0570.1RetroAgent2026.0569ReACT2026.0563.7GiGPO2026.0557.1LATS2026.0554.5rStar2026.0546.4HGPO2026.0544.9