List your company

Mathematical Reasoning on Five-Bench Unweighted Mean

31.06Average ScoreRollout-level |Ai| replay
10.31215.698521.08526.4715Jun 3, 2026

Evaluation results

MethodLinks
Rollout-level |Ai| replayModel=Qwen3-4B2026.0631.0650.98
GRPOModel=Qwen3-4B2026.0626.7140.92
uniformModel=Qwen3-4B2026.0626.742.99
σgModel=Qwen3-4B2026.0626.6946.01
Rollout-level |Ai| replayModel=Qwen3-1.7B2026.0616.5832.35
σgModel=Qwen3-1.7B2026.0616.1228.33
uniformModel=Qwen3-1.7B2026.0615.9327.99
GRPOModel=Qwen3-1.7B2026.0615.7527.64
Rollout-level |Ai| replayModel=Qwen3-0.6B2026.0611.5823.24
σgModel=Qwen3-0.6B2026.0611.3122.57
GRPOModel=Qwen3-0.6B2026.0611.221.32
uniformModel=Qwen3-0.6B2026.0611.1120.01