List your company
ResearchFormerly WizWand

Math Reasoning on Olympiad Bench (Accuracy %)

55.4AccuracyGRPO
16.9226.9136.946.89May 9, 2026May 10, 2026May 11, 2026
Updated 5mo ago

Evaluation results

MethodLinks
GRPOModel=Qwen3-4B-Base, QB (Query-preserved mini batching)=true, RB (Reward-balanced batching)=true2026.0555.4—
GRPOModel=Qwen3-4B-Base, QB (Query-preserved mini batching)=false, RB (Reward-balanced batching)=true2026.0554.8—
GRPOModel=Qwen3-4B-Base, QB (Query-preserved mini batching)=true, RB (Reward-balanced batching)=false2026.0553.6—
GRPOModel=Qwen3-4B-Base, QB (Query-preserved mini batching)=false, RB (Reward-balanced batching)=false2026.0553.2—
ED-iDPOBase Model=Qwen2.5-7B-Instruct-SFT, Self-Consistency=true2026.0552.812.4
ED-GRPOBase Model=Qwen2.5-7B-Instruct-SFT, Self-Consistency=true2026.0552.512.1
GRPOBase Model=Qwen2.5-7B-Instruct-SFT, Self-Consistency=true2026.0552.412
CoTBase Model=Qwen2.5-7B-Instruct-SFT, Self-Consistency=true2026.0552.111.7
DAPOBase Model=Qwen2.5-7B-Instruct-SFT, Self-Consistency=true2026.0551.611.2
ETOBase Model=Qwen2.5-7B-Instruct-SFT, Self-Consistency=true2026.055110.6
GRPOModel=Qwen2.5-Math-7B, QB (Query-preserved mini batching)=true, RB (Reward-balanced batching)=true2026.0544.5—
ED-GRPOBase Model=Qwen2.5-7B-Instruct-SFT, Self-Consistency=false2026.0543.93.5
GRPOBase Model=Qwen2.5-7B-Instruct-SFT, Self-Consistency=false2026.0542.92.5
ETOBase Model=Qwen2.5-7B-Instruct-SFT, Self-Consistency=false2026.0542.72.3
GRPOModel=Qwen2.5-Math-7B, QB (Query-preserved mini batching)=true, RB (Reward-balanced batching)=false2026.0542.6—
ED-iDPOBase Model=Qwen2.5-7B-Instruct-SFT, Self-Consistency=false2026.0542.62.2
DAPOBase Model=Qwen2.5-7B-Instruct-SFT, Self-Consistency=false2026.0541.81.4
CoTBase Model=Qwen2.5-7B-Instruct-SFT, Self-Consistency=false2026.0540.4—
GRPOModel=Qwen2.5-Math-7B, QB (Query-preserved mini batching)=false, RB (Reward-balanced batching)=true2026.0539—
GRPOModel=Qwen3-1.7B-Base, QB (Query-preserved mini batching)=true, RB (Reward-balanced batching)=true2026.0538.2—
GRPOModel=Qwen2.5-Math-7B, QB (Query-preserved mini batching)=false, RB (Reward-balanced batching)=false2026.0537.8—
GRPOModel=Qwen3-1.7B-Base, QB (Query-preserved mini batching)=true, RB (Reward-balanced batching)=false2026.0537.6—
GRPOModel=Qwen3-1.7B-Base, QB (Query-preserved mini batching)=false, RB (Reward-balanced batching)=true2026.0537.3—
GRPOModel=Qwen3-1.7B-Base, QB (Query-preserved mini batching)=false, RB (Reward-balanced batching)=false2026.0534.5—
ED-iDPOBase Model=LLaMA3.1-8B-Instruct-SFT, Self-Consistency=true2026.0530.111.4
ED-GRPOBase Model=LLaMA3.1-8B-Instruct-SFT, Self-Consistency=true2026.0529.510.8
CoTBase Model=LLaMA3.1-8B-Instruct-SFT, Self-Consistency=true2026.0528.29.5
GRPOBase Model=LLaMA3.1-8B-Instruct-SFT, Self-Consistency=true2026.0527.99.2
ETOBase Model=LLaMA3.1-8B-Instruct-SFT, Self-Consistency=true2026.0527.79
DAPOBase Model=LLaMA3.1-8B-Instruct-SFT, Self-Consistency=true2026.0526.17.4
ED-GRPOBase Model=LLaMA3.1-8B-Instruct-SFT, Self-Consistency=false2026.0520.61.9
ED-iDPOBase Model=LLaMA3.1-8B-Instruct-SFT, Self-Consistency=false2026.0519.60.9
ETOBase Model=LLaMA3.1-8B-Instruct-SFT, Self-Consistency=false2026.0519.10.4
GRPOBase Model=LLaMA3.1-8B-Instruct-SFT, Self-Consistency=false2026.0519.10.4
CoTBase Model=LLaMA3.1-8B-Instruct-SFT, Self-Consistency=false2026.0518.7—
DAPOBase Model=LLaMA3.1-8B-Instruct-SFT, Self-Consistency=false2026.0518.4-0.3