List your company

Commonsense Reasoning on LM-Evaluation-Harness Zero-Shot Suite

69.87ARC-Easy AccuracyMamba2
54.280458.327762.37566.4223Jun 7, 2026

Evaluation results

MethodLinks
Mamba2number of parameters=1.3B, training tokens=30B, evaluation mode=zero-shot2026.0669.8736.3553.2440.6870.2956.0455.8137.452.46
Gated DeltaNetnumber of parameters=1.3B, training tokens=30B, evaluation mode=zero-shot2026.0668.633.2852.643.870.8454.7859.4238.852.77
Q-Deltanumber of parameters=1.3B, training tokens=30B, evaluation mode=zero-shot2026.0668.2736.653.4643.2871.4454.9361.4138.453.47
Mambanumber of parameters=1.3B, training tokens=30B, evaluation mode=zero-shot2026.0668.136.1853.4440.7772.255.0155.6337.852.39
DeltaNetnumber of parameters=1.3B, training tokens=30B, evaluation mode=zero-shot2026.0667.6334.4751.0941.7870.9554.761.1938.452.53
RetNetnumber of parameters=1.3B, training tokens=30B, evaluation mode=zero-shot2026.0663.6833.3647.7338.769.0452.7260.6136.650.31
Gated DeltaNetnumber of parameters=340M, training tokens=15B, evaluation mode=zero-shot2026.0660.0225.9440.2531.5267.351.5457.1334.446.01
Mambanumber of parameters=340M, training tokens=15B, evaluation mode=zero-shot2026.0659.6829.1842.9832.9767.5251.7855.8133.246.64
Q-Deltanumber of parameters=340M, training tokens=15B, evaluation mode=zero-shot2026.0659.5128.541.6133.967.6352.8859.4834.447.24
Mamba2number of parameters=340M, training tokens=15B, evaluation mode=zero-shot2026.0659.329.0142.1333.6768.0152.3351.933.846.27
RetNetnumber of parameters=340M, training tokens=15B, evaluation mode=zero-shot2026.0657.0728.4138.7127.3666.5449.4156.883244.55
DeltaNetnumber of parameters=340M, training tokens=15B, evaluation mode=zero-shot2026.0654.8827.4738.6527.1163.649.9659.4629.443.82