ResearchBenchmarksScientific Reasoning on SuperGPQA EngFollow45AccuracyRLCER17.4424.59531.7538.905Feb 11, 2026Evaluation ResultsMethodMethodLinksAccuracyRLCERScale=8BScale=8B2026.0245RLVRScale=8BScale=8B2026.0242.94RLCERScale=4BScale=4B2026.0240.19RLVRScale=4BScale=4B2026.0239.75SFTScale=8BScale=8B2026.0236BaseScale=8BScale=8B2026.0233.06SFTScale=4BScale=4B2026.0231.69BaseScale=4BScale=4B2026.0218.5