List your company
ResearchFormerly WizWand

Math Reasoning on MATH 500 (Pass@4)

96.8Pass@4Qwen3 (ExpertCondenser)
14.95236.20157.4578.699Apr 24, 2026Apr 25, 2026Apr 26, 2026Apr 27, 2026Apr 28, 2026
Updated 6mo ago

Evaluation results

MethodLinks
Qwen3 (ExpertCondenser)Model Size=30B, Distill Type=ExpertCondenser, Fine-tuning with Stanford-S1=true2026.0496.8
Qwen3 (DenseMixer)Model Size=30B, Distill Type=DenseMixer, Fine-tuning with Stanford-S1=true2026.0495.8
Qwen3 (SFT)Model Size=30B, Distill Type=SFT, Fine-tuning with Stanford-S1=true2026.0494.8
Qwen3 (ESFT)Model Size=30B, Distill Type=ESFT, Fine-tuning with Stanford-S1=true2026.0492
JURY-RLBackbone=Qwen2.5-7B2026.0475.65
GT-RewardBackbone=Qwen2.5-7B2026.0475.1
CoRewardBackbone=Qwen2.5-7B2026.0473.85
EntropyBackbone=Qwen2.5-7B2026.0473.15
LLM-KDBackbone=Qwen2.5-7B2026.0472.7
Self-CertaintyBackbone=Qwen2.5-7B2026.0472.45
LLM-as-a-JudgeBackbone=Qwen2.5-7B2026.0472.3
Majority-VotingBackbone=Qwen2.5-7B2026.0471
GT-RewardBackbone=Qwen3-1.7B-Base2026.0468.35
LLM-KDBackbone=Qwen3-1.7B-Base2026.0467.8
DeepSeek-Coder-V2-Lite (ExpertCondenser)Model Size=16B, Distill Type=ExpertCondenser, Fine-tuning with Stanford-S1=true2026.0466.8
JURY-RLBackbone=Qwen3-1.7B-Base2026.0466.4
CoRewardBackbone=Qwen3-1.7B-Base2026.0465.4
EntropyBackbone=Qwen3-1.7B-Base2026.0465
DeepSeek-Coder-V2-Lite (DenseMixer)Model Size=16B, Distill Type=DenseMixer, Fine-tuning with Stanford-S1=true2026.0464.8
DeepSeek-Coder-V2-Lite (SFT)Model Size=16B, Distill Type=SFT, Fine-tuning with Stanford-S1=true2026.0464.6
DeepSeek-Coder-V2-Lite (ESFT)Model Size=16B, Distill Type=ESFT, Fine-tuning with Stanford-S1=true2026.0463
LLM-as-a-JudgeBackbone=Qwen3-1.7B-Base2026.0463
Majority-VotingBackbone=Qwen3-1.7B-Base2026.0459.6
Self-CertaintyBackbone=Qwen3-1.7B-Base2026.0457.05
LLM-KDBackbone=Llama-3.2-3B-Instruct2026.0451.25
Before RLBackbone=Qwen2.5-7B2026.0449.2
JURY-RLBackbone=Llama-3.2-3B-Instruct2026.0448.8
Majority-VotingBackbone=Llama-3.2-3B-Instruct2026.0448.4
GT-RewardBackbone=Llama-3.2-3B-Instruct2026.0447.5
LLM-as-a-JudgeBackbone=Llama-3.2-3B-Instruct2026.0447.35
CoRewardBackbone=Llama-3.2-3B-Instruct2026.0447.2
Before RLBackbone=Qwen3-1.7B-Base2026.0446.85
Before RLBackbone=Llama-3.2-3B-Instruct2026.0444.4
Self-CertaintyBackbone=Llama-3.2-3B-Instruct2026.0442.4
EntropyBackbone=Llama-3.2-3B-Instruct2026.0440.5
Qwen1.5-MoE (ExpertCondenser)Model Size=14B, Distill Type=ExpertCondenser, Fine-tuning with Stanford-S1=true2026.0428.6
Qwen1.5-MoE (DenseMixer)Model Size=14B, Distill Type=DenseMixer, Fine-tuning with Stanford-S1=true2026.0426.7
Qwen1.5-MoE (SFT)Model Size=14B, Distill Type=SFT, Fine-tuning with Stanford-S1=true2026.0420.1
Qwen1.5-MoE (ESFT)Model Size=14B, Distill Type=ESFT, Fine-tuning with Stanford-S1=true2026.0418.1