List your company
ResearchFormerly WizWand

Mathematical Reasoning on Olympiad (Accuracy)

73.2Accuracy4B-Instruct
-2.896816.859136.61556.3709Dec 2, 2025Dec 17, 2025Jan 1, 2026Jan 16, 2026Jan 31, 2026Feb 15, 2026Mar 3, 2026
Updated 8mo ago

Evaluation results

MethodLinks
4B-InstructModel=4B-Instruct, RL Algorithm=HACPO, Heterogeneity Setting=Qwen3-4B + Qwen3-4B-Instruct2026.0373.2
4B-InstructModel=4B-Instruct, RL Algorithm=GSPO, Heterogeneity Setting=Qwen3-4B + Qwen3-4B-Instruct2026.0372.6
4B-InstructModel=4B-Instruct, RL Algorithm=Base, Heterogeneity Setting=Qwen3-4B + Qwen3-4B-Instruct2026.0372.2
4B-InstructModel=4B-Instruct, RL Algorithm=GRPO, Heterogeneity Setting=Qwen3-4B + Qwen3-4B-Instruct2026.0372
STAPOBase Model=Qwen3-14B, Evaluation Setting=JustRL2026.0271.62
4B-InstructModel=4B-Instruct, RL Algorithm=GSPO×2, Heterogeneity Setting=Qwen3-4B + Qwen3-4B-Instruct2026.0371.1
JustRLBase Model=Qwen3-14B, Evaluation Setting=JustRL2026.0268.81
STAPOBase Model=Qwen3-14B, Evaluation Setting=Training-Aligned2026.0268.21
GRPOBase Model=Qwen3-14B, Evaluation Setting=Training-Aligned2026.0264.76
4BModel=4B, RL Algorithm=HACPO, Heterogeneity Setting=Qwen3-4B + Qwen3-4B-Instruct2026.0364.3
GRPOBase Model=Qwen3-14B, Evaluation Setting=JustRL2026.0264.28
20-EntropyBase Model=Qwen3-14B, Evaluation Setting=JustRL2026.0263.95
STAPOBase Model=Qwen3-8B, Evaluation Setting=JustRL2026.0263.32
STAPOBase Model=Qwen3-8B, Evaluation Setting=Training-Aligned2026.0262.98
20-EntropyBase Model=Qwen3-8B, Evaluation Setting=Training-Aligned2026.0262.5
GRPOBase Model=Qwen3-8B, Evaluation Setting=Training-Aligned2026.0261.5
20-EntropyBase Model=Qwen3-14B, Evaluation Setting=Training-Aligned2026.0261.3
20-EntropyBase Model=Qwen3-8B, Evaluation Setting=JustRL2026.0260.39
4BModel=4B, RL Algorithm=GRPO, Heterogeneity Setting=Qwen3-4B + Qwen3-4B-Instruct2026.0359.2
GRPOBase Model=Qwen3-8B, Evaluation Setting=JustRL2026.0258.34
4BModel=4B, RL Algorithm=GSPO×2, Heterogeneity Setting=Qwen3-4B + Qwen3-4B-Instruct2026.0357.9
JustRLBase Model=Qwen3-14B, Evaluation Setting=Training-Aligned2026.0257.01
JustRLBase Model=Qwen3-8B, Evaluation Setting=JustRL2026.0256.57
4BModel=4B, RL Algorithm=GSPO, Heterogeneity Setting=Qwen3-4B + Qwen3-4B-Instruct2026.0356.4
4B-InstructModel=4B-Instruct, RL Algorithm=Naive, Heterogeneity Setting=Qwen3-4B + Qwen3-4B-Instruct2026.0355.2
4BModel=4B, RL Algorithm=Base, Heterogeneity Setting=Qwen3-4B + Qwen3-4B-Instruct2026.0352.4
JustRLBase Model=Qwen3-8B, Evaluation Setting=Training-Aligned2026.0251.26
BAPORollouts=733k, Type=off, Backbone=DeepSeek R1 Distill Qwen 1.5B2026.0250.06
DAPORollouts=1921k, Type=on, Backbone=DeepSeek R1 Distill Qwen 1.5B2026.0248.48
4B-BaseModel=4B-Base, RL Algorithm=HACPO, Heterogeneity Setting=Qwen3-1.7B-Base + Qwen3-4B-Base2026.0346.7
4B-BaseModel=4B-Base, RL Algorithm=GSPO×2, Heterogeneity Setting=Qwen3-1.7B-Base + Qwen3-4B-Base2026.0346.3
qwen3-4BModel=qwen3-4B, RL Algorithm=GSPO×2, Heterogeneity Setting=Qwen3-4B-Base + Llama3.2-3B-Instruct2026.0346.3
GRPO (v = 5)Rollouts=677k, Type=off, Backbone=DeepSeek R1 Distill Qwen 1.5B2026.0246.18
4B-BaseModel=4B-Base, RL Algorithm=GSPO, Heterogeneity Setting=Qwen3-1.7B-Base + Qwen3-4B-Base2026.0346
qwen3-4BModel=qwen3-4B, RL Algorithm=GSPO, Heterogeneity Setting=Qwen3-4B-Base + Llama3.2-3B-Instruct2026.0346
MoPPSRollouts=737k, Type=on, Backbone=DeepSeek R1 Distill Qwen 1.5B2026.0245.93
RePORollouts=677k, Type=off, Backbone=DeepSeek R1 Distill Qwen 1.5B2026.0245.44
4B-BaseModel=4B-Base, RL Algorithm=GRPO, Heterogeneity Setting=Qwen3-1.7B-Base + Qwen3-4B-Base2026.0345.4
qwen3-4BModel=qwen3-4B, RL Algorithm=GRPO, Heterogeneity Setting=Qwen3-4B-Base + Llama3.2-3B-Instruct2026.0345.4
GRPORollouts=677k, Type=on, Backbone=DeepSeek R1 Distill Qwen 1.5B2026.0245.33
DeepSeek R1 Distill Qwen 1.5BRollouts=N/A, Type=N/A2026.0244.42
qwen3-4BModel=qwen3-4B, RL Algorithm=HACPO, Heterogeneity Setting=Qwen3-4B-Base + Llama3.2-3B-Instruct2026.0344.2
Remix-GRPORollouts=N/A, Type=off, Backbone=DeepSeek R1 Distill Qwen 1.5B2026.0243.55
CAPO (GRPO)Backbone=Qwen2.5-7B-Math2025.1239.7
4BModel=4B, RL Algorithm=Naive, Heterogeneity Setting=Qwen3-4B + Qwen3-4B-Instruct2026.0339.4
STAPOBase Model=Qwen3-1.7B, Evaluation Setting=Training-Aligned2026.0238.54
STAPOBase Model=Qwen3-1.7B, Evaluation Setting=JustRL2026.0237.87
CAPO (PPO)Backbone=Qwen2.5-7B-Math2025.1237.8
Reinforce++Backbone=Qwen2.5-7B-Math2025.1237.2
CAPO (Reinforce++)Backbone=Qwen2.5-7B-Math2025.1237.2
GRPOBackbone=Qwen2.5-7B-Math2025.1236.9
RLOOBackbone=Qwen2.5-7B-Math2025.1236
20-EntropyBase Model=Qwen3-1.7B, Evaluation Setting=Training-Aligned2026.0235.65
CAPO (RLOO)Backbone=Qwen2.5-7B-Math2025.1235.6
4B-BaseModel=4B-Base, RL Algorithm=Naive, Heterogeneity Setting=Qwen3-1.7B-Base + Qwen3-4B-Base2026.0335.4
qwen3-4BModel=qwen3-4B, RL Algorithm=Naive, Heterogeneity Setting=Qwen3-4B-Base + Llama3.2-3B-Instruct2026.0335.4
4B-BaseModel=4B-Base, RL Algorithm=Base, Heterogeneity Setting=Qwen3-1.7B-Base + Qwen3-4B-Base2026.0334.7
qwen3-4BModel=qwen3-4B, RL Algorithm=Base, Heterogeneity Setting=Qwen3-4B-Base + Llama3.2-3B-Instruct2026.0334.7
20-EntropyBase Model=Qwen3-1.7B, Evaluation Setting=JustRL2026.0234.35
JustRLBase Model=Qwen3-1.7B, Evaluation Setting=JustRL2026.0234.16
PPOBackbone=Qwen2.5-7B-Math2025.1234.1
CAPO (RLOO)Backbone=Qwen2.5-1.5B-Math2025.1233.2
CAPO (PPO)Backbone=Qwen2.5-1.5B-Math2025.1233
CAPO (GRPO)Backbone=Qwen2.5-1.5B-Math2025.1232.9
RLOOBackbone=Qwen2.5-1.5B-Math2025.1232
Reinforce++Backbone=Qwen2.5-1.5B-Math2025.1231.7
CAPO (Reinforce++)Backbone=Qwen2.5-1.5B-Math2025.1231.6
1.7B-BaseModel=1.7B-Base, RL Algorithm=HACPO, Heterogeneity Setting=Qwen3-1.7B-Base + Qwen3-4B-Base2026.0331.4
GRPOBase Model=Qwen3-1.7B, Evaluation Setting=Training-Aligned2026.0230.64
GRPOBase Model=Qwen3-1.7B, Evaluation Setting=JustRL2026.0230.64
GRPOBackbone=Qwen2.5-1.5B-Math2025.1230.1
1.7B-BaseModel=1.7B-Base, RL Algorithm=GRPO, Heterogeneity Setting=Qwen3-1.7B-Base + Qwen3-4B-Base2026.0329.8
1.7B-BaseModel=1.7B-Base, RL Algorithm=GSPO×2, Heterogeneity Setting=Qwen3-1.7B-Base + Qwen3-4B-Base2026.0329.3
JustRLBase Model=Qwen3-1.7B, Evaluation Setting=Training-Aligned2026.0229.15
1.7B-BaseModel=1.7B-Base, RL Algorithm=GSPO, Heterogeneity Setting=Qwen3-1.7B-Base + Qwen3-4B-Base2026.0328.7
CoTBackbone=Qwen2.5-7B-Math2025.1227.8
CoTBackbone=Qwen2.5-1.5B-Math2025.1227.6
PPOBackbone=Qwen2.5-1.5B-Math2025.1227.1
1.7B-BaseModel=1.7B-Base, RL Algorithm=Naive, Heterogeneity Setting=Qwen3-1.7B-Base + Qwen3-4B-Base2026.0326.3
TARSEs2026.0324.7
1.7B-BaseModel=1.7B-Base, RL Algorithm=Base, Heterogeneity Setting=Qwen3-1.7B-Base + Qwen3-4B-Base2026.0322.9
llama3.2-3BModel=llama3.2-3B, RL Algorithm=HACPO, Heterogeneity Setting=Qwen3-4B-Base + Llama3.2-3B-Instruct2026.0320.8
rStar-Qwen2Backbone=Qwen22026.0320
llama3.2-3BModel=llama3.2-3B, RL Algorithm=GRPO, Heterogeneity Setting=Qwen3-4B-Base + Llama3.2-3B-Instruct2026.0317.4
llama3.2-3BModel=llama3.2-3B, RL Algorithm=GSPO, Heterogeneity Setting=Qwen3-4B-Base + Llama3.2-3B-Instruct2026.0317
llama3.2-3BModel=llama3.2-3B, RL Algorithm=GSPO×2, Heterogeneity Setting=Qwen3-4B-Base + Llama3.2-3B-Instruct2026.0315.9
llama3.2-3BModel=llama3.2-3B, RL Algorithm=Base, Heterogeneity Setting=Qwen3-4B-Base + Llama3.2-3B-Instruct2026.0315.8
llama3.2-3BModel=llama3.2-3B, RL Algorithm=Naive, Heterogeneity Setting=Qwen3-4B-Base + Llama3.2-3B-Instruct2026.0310.7
Qwen2-7B-InstructType=Instruct, Parameters=7B2026.030.04
NuminaMath-7B-COTType=Chain-of-Thought, Parameters=7B2026.030.03