List your company

Mathematical Reasoning on AIME 2025 (Pass@1, FS, N/T, D/T, Repro)

72.2Pass@1 AccuracyRED
44.53651.71858.966.082Apr 3, 2026

Evaluation results

MethodLinks
REDBackbone=Qwen3-32B-thinking2026.0472.23.14.23.10.026
S-GRPOBackbone=Qwen3-32B-thinking2026.0471.16.175.10.074
VanillaBackbone=Qwen3-32B-thinking2026.0468.96.87.45.30.081
DEERBackbone=Qwen3-32B-thinking2026.0466.76.37.25.40.083
RL + LPBackbone=Qwen3-32B-thinking2026.0466.75.96.94.70.071
REDBackbone=R1-Qwen-32B2026.0463.33.24.63.60.04
RL + LPBackbone=R1-Qwen-32B2026.0462.26.17.35.20.085
S-GRPOBackbone=R1-Qwen-32B2026.04606.37.45.60.088
VanillaBackbone=R1-Qwen-32B2026.0458.977.85.80.095
DEERBackbone=R1-Qwen-32B2026.0457.86.57.65.90.097
REDBackbone=R1-Llama-70B2026.04503.94.73.70.04
DEERBackbone=R1-Llama-70B2026.0448.97.97.86.50.128
S-GRPOBackbone=R1-Llama-70B2026.0448.97.87.66.10.115
VanillaBackbone=R1-Llama-70B2026.0447.88.68.36.40.125
RL + LPBackbone=R1-Llama-70B2026.0445.67.67.55.80.11