List your company

Mathematical Reasoning on MultiArith (Acc, #L, Acc/#L)

95AccuracyRuPLaR-Mix
16.16836.63457.177.566May 10, 2026

Evaluation results

MethodLinks
RuPLaR-Mix2026.05952.0346.8
RuPLaR-Gumbel2026.0593.72.0246.4
RuPLaRPrior Strength (alpha_kl)=0.32026.0591.51.9946
ReGuLaR2026.0589.82.0439.1
RuPLaR-Temp2026.0589.82.0444
RuPLaRLoss Component=w/o KL-Focus2026.0589.52.0643.4
RuPLaRPrior Strength (alpha_kl)=2.02026.0589.22.0144.4
RuPLaRLoss Component=w/o Problem-Thought2026.0588.82.0443.5
CoLaR-52026.0586.43.2126.9
RuPLaRPrior Type=Learned2026.0582.32.0739.8
RuPLaRPrior Type=Random2026.05811.9840.9
RuPLaRPrior Type=Uniform2026.0572.22.0335.6
COCONUT2026.0541.166.85
iCoT2026.0538.20—
CODI2026.0519.263.2