List your company

Mathematical Reasoning on GSM Hard (Accuracy, Line Count, Accuracy per Line)

10.8AccuracyRuPLaR-Mix
2.65684.77096.8858.9991May 10, 2026

Evaluation results

MethodLinks
RuPLaR-Mix2026.0510.82.873.76
RuPLaR-Gumbel2026.0510.22.883.54
RuPLaR-Temp2026.059.862.783.55
RuPLaRPrior Type=Learned2026.059.782.883.4
RuPLaRLoss Component=w/o Problem-Thought2026.059.72.763.51
RuPLaRPrior Strength (alpha_kl)=2.02026.059.472.873.3
RuPLaRLoss Component=w/o KL-Focus2026.059.42.763.41
RuPLaRPrior Strength (alpha_kl)=0.32026.059.332.623.56
ReGuLaR2026.058.274.122.01
RuPLaRPrior Type=Random2026.058.262.653.12
RuPLaRPrior Type=Uniform2026.058.043.242.48
CoLaR-52026.055.876.530.9
COCONUT2026.055.4960.92
iCoT2026.053.870—
CODI2026.052.9760.5