List your company

Mathematical Problem Solving on Aggregated OOD Math Benchmarks (MATH500, AIME24, Minerva-Math, AMC23)

91.8MATH500 ScoreDFPO
84.10486.10288.190.098Feb 5, 2026

Evaluation results

MethodLinks
DFPOtraining_domain=Science2026.0291.856.6732.3588.3367.2940.25
λ-GRPOtraining_domain=Science2026.0289.643.3329.0482.561.1236.45
FlowRLtraining_domain=Science2026.0289.65031.6286.6764.4738.37
Reinforce++training_domain=Science2026.0289.45030.5183.3363.3137.72
Dr. GRPOtraining_domain=Science2026.0288.648.3327.218061.0436.25
Basetraining_domain=Science2026.0288.45031.2580.8362.6237.18
Robust Bellmantraining_domain=Science2026.0287.44527.2183.3360.7436.23
KTAEtraining_domain=Science2026.0287.246.6730.1582.561.6336.62
PPOtraining_domain=Science2026.028751.6729.418563.2737.34
GRPOtraining_domain=Science2026.0286.84526.8482.560.2935.77
BAPOtraining_domain=Science2026.0284.46026.8482.563.4437.71