ResearchBenchmarksGeneral Capability Evaluation on Average (MMLU, GSM8K, MBPP)Follow78.84AccuracyBaseline71.653673.519375.38577.2507Mar 16, 2026Evaluation ResultsMethodMethodLinksAccuracyUtility PreservationBaseline2026.0378.84—SFCoT2026.0371.9391.2