List your company
ResearchFormerly WizWand

Mathematical Reasoning on OlympiadBench (Pass Rate)

60.67Pass RatePREPO
7.557221.346135.13548.9239Nov 2, 2025Nov 18, 2025Dec 4, 2025Dec 20, 2025Jan 5, 2026Jan 21, 2026Feb 7, 2026
Updated 8mo ago

Evaluation results

MethodLinks
PREPOBase Model=Qwen3-4B, Training Strategy=PREPO, # Rollouts=348K2025.1160.67
Random SelectionBase Model=Qwen3-4B, Training Strategy=Random Selection, # Rollouts=553K2025.1159.33
GRESOBase Model=Qwen3-4B, Training Strategy=GRESO, # Rollouts=472K2025.1157.33
Qwen3-4BBase Model=Qwen3-4B, Training Strategy=Base, # Rollouts=–2025.1152.67
SED-SFTBackbone=Qwen2.5-Math-7B-Instruct, Training Phase=RL, Optimization Objective=SED-SFT2026.0250.2
GEMBackbone=Qwen2.5-Math-7B-Instruct, Training Phase=RL, Optimization Objective=GEM2026.0248.9
SED-SFT w/o maskBackbone=Qwen2.5-Math-7B-Instruct, Training Phase=RL, Optimization Objective=SED-SFT w/o mask2026.0248.3
CrossEntropyBackbone=Qwen2.5-Math-7B-Instruct, Training Phase=RL, Optimization Objective=CrossEntropy2026.0246.5
DFTBackbone=Qwen2.5-Math-7B-Instruct, Training Phase=RL, Optimization Objective=DFT2026.0245.8
DFTBackbone=Qwen2.5-Math-7B-Instruct, Training Phase=SFT, Optimization Objective=DFT2026.0244
Random SelectionBase Model=Qwen2.5-Math-7B, Training Strategy=Random Selection, # Rollouts=905K2025.1143.26
GRESOBase Model=Qwen2.5-7B, Training Strategy=GRESO, # Rollouts=680K2025.1142.07
PREPOBase Model=Qwen2.5-Math-7B, Training Strategy=PREPO, # Rollouts=540K2025.1141.58
PREPOBase Model=Qwen2.5-7B, Training Strategy=PREPO, # Rollouts=304K2025.1139.85
Qwen2.5-Math-7BBase Model=Qwen2.5-Math-7B, Training Strategy=Base, # Rollouts=–2025.1139.56
Random SelectionBase Model=Qwen2.5-7B, Training Strategy=Random Selection, # Rollouts=716K2025.1138.47
Qwen2.5-7BBase Model=Qwen2.5-7B, Training Strategy=Base, # Rollouts=–2025.1133.09
PREPOBase Model=Qwen2.5-Math-1.5B, Training Strategy=PREPO, # Rollouts=1.1M2025.1132
Random SelectionBase Model=Qwen2.5-Math-1.5B, Training Strategy=Random Selection, # Rollouts=3.0M2025.1130.5
GEMBackbone=Qwen2.5-Math-7B-Instruct, Training Phase=SFT, Optimization Objective=GEM2026.0228
SED-SFT w/o maskBackbone=Qwen2.5-Math-7B-Instruct, Training Phase=SFT, Optimization Objective=SED-SFT w/o mask2026.0227.7
Qwen2.5-Math-1.5BBase Model=Qwen2.5-Math-1.5B, Training Strategy=Base, # Rollouts=–2025.1127.41
GRESOBase Model=Qwen2.5-Math-7B, Training Strategy=GRESO, # Rollouts=654K2025.1126.83
SED-SFTBackbone=Qwen2.5-Math-7B-Instruct, Training Phase=SFT, Optimization Objective=SED-SFT2026.0226.5
CrossEntropyBackbone=Qwen2.5-Math-7B-Instruct, Training Phase=SFT, Optimization Objective=CrossEntropy2026.0226.1
GRESOBase Model=Qwen2.5-Math-1.5B, Training Strategy=GRESO, # Rollouts=2.5M2025.1124.17
SED-SFTBackbone=Llama-3.2-3B-Instruct, Training Phase=RL, Optimization Objective=SED-SFT2026.0222.2
SED-SFT w/o maskBackbone=Llama-3.2-3B-Instruct, Training Phase=RL, Optimization Objective=SED-SFT w/o mask2026.0221
CrossEntropyBackbone=Llama-3.2-3B-Instruct, Training Phase=RL, Optimization Objective=CrossEntropy2026.0220.9
DFTBackbone=Llama-3.2-3B-Instruct, Training Phase=RL, Optimization Objective=DFT2026.0217.8
GEMBackbone=Llama-3.2-3B-Instruct, Training Phase=RL, Optimization Objective=GEM2026.0216.9
DFTBackbone=Llama-3.2-3B-Instruct, Training Phase=SFT, Optimization Objective=DFT2026.0212.1
SED-SFT w/o maskBackbone=Llama-3.2-3B-Instruct, Training Phase=SFT, Optimization Objective=SED-SFT w/o mask2026.0210.7
GEMBackbone=Llama-3.2-3B-Instruct, Training Phase=SFT, Optimization Objective=GEM2026.029.9
CrossEntropyBackbone=Llama-3.2-3B-Instruct, Training Phase=SFT, Optimization Objective=CrossEntropy2026.029.6
SED-SFTBackbone=Llama-3.2-3B-Instruct, Training Phase=SFT, Optimization Objective=SED-SFT2026.029.6