List your company

Mathematical Reasoning on Minerva-Math (avg@1)

47.1Avg@1 AccuracyNPR (Variant)
8.817618.756328.69538.6337Dec 3, 2025Dec 17, 2025Jan 1, 2026Jan 16, 2026Jan 30, 2026Feb 14, 2026Mar 1, 2026

Evaluation results

MethodLinks
NPR (Variant)Data=orz-8k, Train=Parallel RL, Base=NPR-BETA2025.1247.1
NPR-BETA (Variant)Data=orz-8k, Train=Parallel SFT, Base=Q3-4B2025.1245.9
NPRData=orz-8k, Train=Parallel RL, Base=NPR-BETA2025.1243
SR-BETAData=orz-8k, Train=Sequential SFT, Base=Q3-4B-Inst.2025.1241.5
Qwen3-4B-Instruct-2507Data=N/A, Train=N/A, Base=N/A2025.1241.2
NPR-BETAData=orz-8k, Train=Parallel SFT, Base=Q3-4B-Inst.2025.1241.2
Qwen2.5-32B-InstructData=N/A, Train=N/A, Base=N/A2025.1240.8
Multiverse-32BData=s1.1-8k, Train=S→P SFT, Base=Q2.5-32B-Inst.2025.1240
SRData=orz-8k, Train=Sequential RL, Base=NPR-BETA2025.1238.2
Multiverse-4BData=s1.1-8k, Train=S→P SFT, Base=Q3-4B-Inst.2025.1234.9
Robust BellmanTraining Domain=Math Domain2025.1231.99
DVPOTraining Domain=Math Domain2025.1231.62
PPOTraining Domain=Math Domain2025.1230.51
BaseTraining Domain=Math Domain2025.1228.68
GRPOTraining Domain=Math Domain2025.1228.68
Qwen3-4B (Non-Thinking)Data=N/A, Train=N/A, Base=N/A2025.1228.5
Dr.GRPOTraining Domain=Math Domain2025.1227.94
Reinforce++Training Domain=Math Domain2025.1227.21
LCO-LCHBackbone=Qwen-3-4B2026.0324.26
LCO-KLDBackbone=Qwen-3-4B2026.0323.95
GRPOBackbone=Qwen-3-4B2026.0321.75
LCO-MSEBackbone=Qwen-3-4B2026.0321.37
PPOBackbone=Qwen-3-4B2026.0320.95
DAPOBackbone=Qwen-3-4B2026.0319.75
REINFORCEBackbone=Qwen-3-4B2026.0319.48
GSPOBackbone=Qwen-3-4B2026.0318.38
LCO-KLDBackbone=Qwen-2.5-3B2026.0316.71
ϕDPOBackbone=Qwen-3-4B2026.0316.54
ϕDPOBackbone=Qwen-2.5-3B2026.0316.17
LCO-LCHBackbone=Qwen-2.5-3B2026.0315.38
π*Backbone=Qwen-2.5-3B2026.0315.18
LCO-MSEBackbone=Qwen-2.5-3B2026.0315.12
GRPOBackbone=Qwen-2.5-3B2026.0314.89
GSPOBackbone=Qwen-2.5-3B2026.0314.87
π*Backbone=Qwen-3-4B2026.0314.39
PPOBackbone=Qwen-2.5-3B2026.0313.97
DAPOBackbone=Qwen-2.5-3B2026.0313.23
SFTBackbone=Qwen-3-4B2026.0312.53
REINFORCEBackbone=Qwen-2.5-3B2026.0312.35
SFTBackbone=Qwen-2.5-3B2026.0310.29