ResearchBenchmarksScience Reasoning on GPQA (Acc, PCE, ECE)Follow37.1AccuracyRLCM16.50821.85427.232.546Apr 25, 2026Evaluation ResultsMethodMethodLinksAccuracyPCEECERLCMConf.=probeConf.=probe2026.0437.10.0430.044GRPOConf.=probeConf.=probe2026.04360.050.054C^2GSPGConf.=logitsConf.=logits2026.0432.60.4710.471BaseConf.=probeConf.=probe2026.04280.2630.263RLCRConf.=verbalConf.=verbal2026.0417.30.5170.517