List your company

Zero-shot Commonsense Reasoning Suite (PIQA, SIQA, HellaSwag, ARC-E, ARC-C, WinoGrande, LAMBADA)

66.92PIQA AccuracyReMoE
57.87260.22162.5764.919May 11, 2026

Evaluation results

MethodLinks
ReMoEModel Scale=Large, Routing Strategy=NG2026.0566.9238.7932.6655.9324.5751.4127.142.48
DenseModel Scale=Large2026.0566.8737.7732.9256.2724.7451.729.3442.8
DeepSeek-V3Model Scale=Large, Routing Strategy=GA2026.0566.3238.3832.7155.1824.9151.0728.0642.38
DECOModel Scale=Large, Routing Strategy=GA2026.0566.1438.7432.6956.4123.7251.7826.5742.29
BlockFFNModel Scale=Large2026.0566.138.5932.5156.3624.1551.3827.342.34
ReMoEModel Scale=Large, Routing Strategy=GA2026.0566.0538.6932.0855.2624.3252.3327.1942.27
DeepSeek-V3Model Scale=Large, Routing Strategy=NG2026.0565.9437.8732.7554.6723.3852.4927.0342.02
DECOModel Scale=Large, Routing Strategy=NG2026.0565.9439.6132.7857.1125.4353.2825.542.81
TopPModel Scale=Large2026.0564.5336.4431.4250.5522.5351.0714.1338.67
DECOModel Scale=Small, MoE Normalization Variant=NG2026.0560.9436.8527.9645.8420.2250.6715.836.9
BlockFFNModel Scale=Small2026.0560.8336.2827.6244.8719.4550.1216.5536.53
DenseModel Scale=Small2026.0560.7735.5728.0345.8820.2250.5916.5136.8
DeepSeek-V3Model Scale=Small, MoE Normalization Variant=NG2026.0560.6636.6327.7645.7620.0550.4315.8536.73
ReMoEModel Scale=Small, MoE Normalization Variant=GA2026.0560.3436.5927.7246.3418.4351.2215.636.61
TopPModel Scale=Small2026.0560.1735.2127.6742.3818.8650.917.7434.71
DeepSeek-V3Model Scale=Small, MoE Normalization Variant=GA2026.0560.1235.9327.8544.1519.4550.7515.8936.31
ReMoEModel Scale=Small, MoE Normalization Variant=NG2026.0559.7435.1627.9646.0919.5449.9617.7636.6
DECOModel Scale=Small, MoE Normalization Variant=GA2026.0558.2235.9327.6542.5119.0351.315.8435.78