List your company
ResearchFormerly WizWand

Mathematical Reasoning on AMC23 (Average@16 and Avg.)

91.09Average@16Teacher
11.7932.377552.96573.5525Oct 1, 2025Nov 9, 2025Dec 19, 2025Jan 28, 2026Mar 8, 2026Apr 17, 2026May 27, 2026
Updated 5mo ago

Evaluation results

MethodLinks
TeacherBackbone=QwQ-32B, Prompting=zero-shot2026.0491.0973.9
IRDSModel=QWEN3-4B, Data Budget=30%2026.0582.4—
IRDSModel=QWEN3-4B, Data Budget=10%2026.0576.8—
PPL-TOPModel=QWEN3-4B, Data Budget=30%2026.0574.6—
DEPOModel=QWEN3-4B, Data Budget=30%2026.0571.4—
LIMRModel=QWEN3-4B, Data Budget=30%2026.0571.4—
DenoiseRL-DAPOBase Model=Qwen3-8B-Base2026.0571.4—
PPL-MIDDLEModel=QWEN3-4B, Data Budget=30%2026.0570.6—
DenoiseRL-GRPOBase Model=Qwen3-8B-Base2026.0570.3—
IFDModel=QWEN3-4B, Data Budget=30%2026.0570.2—
GRPOBase Model=Qwen3-8B-Base2026.0569.7—
DAPOBase Model=Qwen3-8B-Base2026.0569.7—
RANDOMModel=QWEN3-4B, Data Budget=30%2026.0564.9—
+CurES-GRPOBase Model=Qwen2.5-Math-7B, Training Method=CurES-GRPO2025.1064.3852.41
+GVM-GRPOBase Model=Qwen2.5-Math-7B, Training Method=GVM-GRPO2025.1064.3150.64
DenoiseRL-DAPOBase Model=Qwen3-4B-Base2026.0563.6—
GRPOBase Model=Qwen3-4B-Base2026.0563.1—
+RPPBase Model=Qwen2.5-Math-7B, Training Method=RPP2025.1062.8150.18
+Speed-RL-RPPBase Model=Qwen2.5-Math-7B, Training Method=Speed-RL-RPP2025.1062.8149.56
DAPOBase Model=Qwen3-4B-Base2026.0562.5—
DenoiseRL-GRPOBase Model=Qwen3-4B-Base2026.0561.4—
+Speed-RL-GRPOBase Model=Qwen2.5-Math-7B, Training Method=Speed-RL-GRPO2025.1060.1648.84
+CurES-RPPBase Model=Qwen2.5-Math-7B, Training Method=CurES-RPP2025.1058.7549.62
+GRPOBase Model=Qwen2.5-Math-7B, Training Method=GRPO2025.1057.547.59
IRDSModel=QWEN3-1.7B, Data Budget=30%2026.0557.5—
TOKEN-LENGTHModel=QWEN3-4B, Data Budget=30%2026.0557.4—
+GVM-RPPBase Model=Qwen2.5-Math-7B, Training Method=GVM-RPP2025.1053.4446.95
IRDSModel=QWEN3-1.7B, Data Budget=10%2026.0552.9—
+CurES-GRPOBase Model=Qwen2.5-Math-1.5B, Training Method=CurES-GRPO2025.1052.1944.94
RANDOMModel=QWEN3-1.7B, Data Budget=30%2026.0551.2—
+GVM-RPPBase Model=Qwen2.5-Math-1.5B, Training Method=GVM-RPP2025.1050.9442.21
+CurES-RPPBase Model=Qwen2.5-Math-1.5B, Training Method=CurES-RPP2025.1050.9444.14
+GVM-GRPOBase Model=Qwen2.5-Math-1.5B, Training Method=GVM-GRPO2025.1050.4742.82
DEPOModel=QWEN3-1.7B, Data Budget=30%2026.0550.2—
BaseBase Model=Qwen3-8B-Base2026.0549.2—
+Speed-RL-RPPBase Model=Qwen2.5-Math-1.5B, Training Method=Speed-RL-RPP2025.1047.3441.75
+Speed-RL-GRPOBase Model=Qwen2.5-Math-1.5B, Training Method=Speed-RL-GRPO2025.1047.1941.34
PPL-TOPModel=QWEN3-1.7B, Data Budget=30%2026.0546.9—
LIMRModel=QWEN3-1.7B, Data Budget=30%2026.0546.4—
+GRPOBase Model=Qwen2.5-Math-1.5B, Training Method=GRPO2025.1045.4741.64
PPL-MIDDLEModel=QWEN3-1.7B, Data Budget=30%2026.0544—
IFDModel=QWEN3-1.7B, Data Budget=30%2026.0543.4—
BaseBase Model=Qwen3-4B-Base2026.0543.1—
Gen-SSDBackbone=Qwen2.5-Math-1.5B, Prompting=zero-shot, Chunk size=4K2026.0441.8834.94
TOKEN-LENGTHModel=QWEN3-1.7B, Data Budget=30%2026.0539.2—
+RPPBase Model=Qwen2.5-Math-1.5B, Training Method=RPP2025.1039.0635.86
MoRSDBackbone=Qwen2.5-Math-1.5B, Prompting=zero-shot2026.0438.7533.07
MCC-KDBackbone=Qwen2.5-Math-1.5B, Prompting=zero-shot2026.0435.4730.25
Standard KDBackbone=Qwen2.5-Math-1.5B, Prompting=zero-shot2026.043529.01
Self-DistillationBackbone=Qwen2.5-Math-1.5B, Prompting=zero-shot2026.0429.5324.25
Qwen2.5-Math-7BBase Model=Qwen2.5-Math-7B, Training Method=Base2025.1027.1933.48
TOKEN-LENGTHModel=LLAMA-3.1-8B-INSTRUCT, Data Budget=30%2026.0523.8—
LIMRModel=LLAMA-3.1-8B-INSTRUCT, Data Budget=30%2026.0523.8—
RANDOMModel=LLAMA-3.1-8B-INSTRUCT, Data Budget=30%2026.0523.2—
DEPOModel=LLAMA-3.1-8B-INSTRUCT, Data Budget=30%2026.0523—
PPL-TOPModel=LLAMA-3.1-8B-INSTRUCT, Data Budget=30%2026.0522.9—
IRDSModel=LLAMA-3.1-8B-INSTRUCT, Data Budget=30%2026.0522.3—
IRDSModel=LLAMA-3.1-8B-INSTRUCT, Data Budget=10%2026.0521—
PPL-MIDDLEModel=LLAMA-3.1-8B-INSTRUCT, Data Budget=30%2026.0520.6—
IFDModel=LLAMA-3.1-8B-INSTRUCT, Data Budget=30%2026.0520.2—
Cold StartBackbone=Qwen2.5-Math-1.5B, Prompting=zero-shot2026.0419.5318.3
StudentBackbone=Qwen2.5-Math-1.5B, Prompting=two-shot2026.0419.0615.54
Qwen2.5-Math-1.5BBase Model=Qwen2.5-Math-1.5B, Training Method=Base2025.1014.8420