List your company
ResearchFormerly WizWand

Mathematical Reasoning on Polaris (test)

4.92Cumulative Rollouts (M)DAPO
0.2921.49352.6953.8965May 26, 2026
Updated 5mo ago

Evaluation results

MethodLinks
DAPOModel Scale=14B, Budget Type=ample, Rollout count per prompt (n)=64, Target Accuracy=57.0%*2026.054.92——
GRPOModel Scale=14B, Budget Type=ample, Rollout count per prompt (n)=64, Target Accuracy=57.0%*2026.054.1——
DAPOModel Scale=8B, Budget Type=ample, Rollout count per prompt (n)=64, Target Accuracy=47.3%2026.052.7——
GRPOModel Scale=4B, Budget Type=ample, Rollout count per prompt (n)=64, Target Accuracy=47.3%*2026.052.5——
Pilot-CommitModel Scale=14B, Budget Type=ample, Rollout count per prompt (n)=64, Target Accuracy=57.0%*2026.052.151.92.3
DAPOModel Scale=4B, Budget Type=limited, Rollout count per prompt (n)=16, Target Accuracy=45.0%2026.051.69——
Pilot-CommitModel Scale=4B, Budget Type=ample, Rollout count per prompt (n)=64, Target Accuracy=47.3%*2026.051.661.5—
GRPOModel Scale=8B, Budget Type=ample, Rollout count per prompt (n)=64, Target Accuracy=47.3%2026.051.56——
DAPOModel Scale=8B, Budget Type=limited, Rollout count per prompt (n)=16, Target Accuracy=47.1%2026.051.26——
Pilot-CommitModel Scale=8B, Budget Type=ample, Rollout count per prompt (n)=64, Target Accuracy=47.3%2026.051.041.52.6
GRPOModel Scale=4B, Budget Type=limited, Rollout count per prompt (n)=16, Target Accuracy=45.0%2026.050.73——
GRPOModel Scale=8B, Budget Type=limited, Rollout count per prompt (n)=16, Target Accuracy=47.1%2026.050.63——
GRPOModel Scale=14B, Budget Type=limited, Rollout count per prompt (n)=16, Target Accuracy=54.3%2026.050.52——
Pilot-CommitModel Scale=4B, Budget Type=limited, Rollout count per prompt (n)=16, Target Accuracy=45.0%2026.050.511.43.3
Pilot-CommitModel Scale=14B, Budget Type=limited, Rollout count per prompt (n)=16, Target Accuracy=54.3%2026.050.511—
Pilot-CommitModel Scale=8B, Budget Type=limited, Rollout count per prompt (n)=16, Target Accuracy=47.1%2026.050.471.32.7