ResearchBenchmarksZero-shot Question Answering on Reasoning Suite (ARC-C, ARC-E, BoolQ, HellaSwag, PIQA, WinoGrande)Follow60.15ARC-C AccuracyFP1651.278853.581955.88558.1881Jun 9, 2026Evaluation ResultsMethodMethodLinksARC-C AccuracyARC-E AccuracyBoolQ AccuracyHellaSwag AccuracyPIQA AccuracyWinoGrande AccuracyAverage Accuracy (Overall)Performance RatioFP16Model Scale=14B, Calib...Model Scale=14B, Calibration Tokens=58M2026.0660.1582.8389.378.8479.8772.8577.311LC-QATModel Scale=14B, Calib...Model Scale=14B, Calibration Tokens=58M2026.0654.6979.2585.973.6678.5173.8874.320.96LC-PTQModel Scale=14B, Calib...Model Scale=14B, Calibration Tokens=58M2026.0651.6277.6987.1669.8277.4271.5972.550.94