List your company

Zero-shot Evaluation on ARC-e, BoolQ, HellaSwag, LAMBADA, PIQA, RACE, SocialIQA, SciQ, and SWAG

77.9ARC-e AccuracyJREG
53.9860.1966.472.61Jan 26, 2026

Evaluation results

MethodLinks
JREGSize=3.4B, alpha=1.02026.0177.961.650.357.275.237.842.59351.560.8
BaselineSize=3.4B, alpha=N/A2026.0175.659.449.755.976.536.343.993.451.360.2
JREGSize=1B, alpha=3.02026.0170.659.342.645.671.834.941.491.247.356.1
JREGSize=1B, alpha=1.02026.0169.461.443.147.271.835.342.188.647.656.2
BaselineSize=1B, alpha=N/A2026.0168.561.442.946.672.432.94189.547.355.8
JREGSize=170M, alpha=1.02026.0157.26032.131.765.229.938.881.140.248.5
JREGSize=170M, alpha=3.02026.0157.26032.431.665.128.94081.739.948.5
JREGSize=170M, alpha=0.52026.015757.132.530.76529.338.682.24048
JREGSize=170M, alpha=0.12026.0156.160.532.229.865.629.639.280.939.948.2
JREGSize=170M, alpha=0.32026.015659.232.231.76529.138.182.24048
JREGSize=170M, alpha=0.02026.0155.758.332.432.565.929.638.681.340.248.3
BaselineSize=170M, alpha=N/A2026.0154.957.532.1326429.138.480.239.747.5