ResearchBenchmarksMathematical Reasoning on MATH-500 Level 5 (Calibration Metrics)Follow47.2Error RateSBBT46.24452.69759.1565.603May 26, 2026Evaluation ResultsMethodMethodLinksError RateObservation ScoreAUROC Gap ScoreBrier DeltaHC AUROC GapText/Self AUROC GapLT/Act AUROC GapSBBTBackbone=Qwen3-14B, Ev...Backbone=Qwen3-14B, Evaluation setting=Stress-test2026.0547.217.2-0.024-0.123-0.023-0.021—SBBTBackbone=7B, Evaluatio...Backbone=7B, Evaluation setting=Main2026.0571.140.70.017-0.1560.0620.110.067