List your company

Mathematical Reasoning on AIME 2025 (Avg@10 accuracy)

36Avg@10 AccuracyAsymGRPO w/ Clip-higher
19.3623.682832.32Apr 6, 2026

Evaluation results

MethodLinks
AsymGRPO w/ Clip-higherBase Model=Qwen3-4B, Temperature=0.4, Top-p=1.02026.0436
GRPO w/ Clip-higherBase Model=Qwen3-4B, Temperature=0.4, Top-p=1.02026.0432.33
Pass@K TrainingBase Model=Qwen3-4B, Temperature=0.4, Top-p=1.02026.0431
Dr.GRPOBase Model=Qwen3-4B, Temperature=0.4, Top-p=1.02026.0430
AsymGRPO (βpos = βneg)Base Model=Qwen3-4B, Temperature=0.4, Top-p=1.02026.0429.33
AsymGRPOBase Model=Qwen3-4B, Temperature=0.4, Top-p=1.02026.0428.67
GRPO w/ Entro.RegularizationBase Model=Qwen3-4B, Temperature=0.4, Top-p=1.02026.0428.33
Pos-Only Modulation (§ 3.1)Base Model=Qwen3-4B, Temperature=0.4, Top-p=1.02026.0428
GRPOBase Model=Qwen3-4B, Temperature=0.4, Top-p=1.02026.0427.33
Neg-Only Modulation (§ 3.1)Base Model=Qwen3-4B, Temperature=0.4, Top-p=1.02026.0427
GRPO w/ Entro.AdvBase Model=Qwen3-4B, Temperature=0.4, Top-p=1.02026.0425.33
REINFORCEBase Model=Qwen3-4B, Temperature=0.4, Top-p=1.02026.0424.67
EntDecrease (§ 3.2)Base Model=Qwen3-4B, Temperature=0.4, Top-p=1.02026.0423.67
EntIncrease (§ 3.2)Base Model=Qwen3-4B, Temperature=0.4, Top-p=1.02026.0423.33
Qwen3-4BTemperature=0.4, Top-p=1.02026.0420