ResearchBenchmarksMathematical Reasoning on AMC23 (Pass@1 accuracy)Follow70Pass@1 AccuracyPUM+GRPO51.856.52561.2565.975Jun 5, 2026Evaluation ResultsMethodMethodLinksPass@1 AccuracyPUM+GRPOSetting=Outcome-AnchoredSetting=Outcome-Anchored2026.0670PUMSetting=Process-OnlySetting=Process-Only2026.0665PRM+GRPOSetting=Outcome-AnchoredSetting=Outcome-Anchored2026.0660PQM+GRPOSetting=Outcome-AnchoredSetting=Outcome-Anchored2026.0660PRMSetting=Process-OnlySetting=Process-Only2026.0657.5PQMSetting=Process-OnlySetting=Process-Only2026.0657.5Vanilla GRPOSetting=Outcome-AnchoredSetting=Outcome-Anchored2026.0652.5