ResearchBenchmarksScientific Reasoning on SuperGPQA SciFollow50.25AccuracyRLCER14.047623.446332.84542.2437Feb 11, 2026Evaluation ResultsMethodMethodLinksAccuracyRLCERScale=8BScale=8B2026.0250.25RLVRScale=8BScale=8B2026.0248.81RLVRScale=4BScale=4B2026.0242.88RLCERScale=4BScale=4B2026.0241.81SFTScale=8BScale=8B2026.0235.19SFTScale=4BScale=4B2026.0227.81BaseScale=8BScale=8B2026.0223.63BaseScale=4BScale=4B2026.0215.44