List your company

Multi-task Language Understanding on MMLU (Vanilla vs Attack Performance)

83.01MMLU Accuracy (Vanilla)AutoRAS
69.406872.938476.4780.0016Jun 19, 2026

Evaluation results

MethodLinks
AutoRASEvaluation Protocol=Vanilla, Cost ($)=0.672026.0683.01—
Magentic-OneEvaluation Protocol=Vanilla, Cost ($)=0.272026.0681.04—
OWLEvaluation Protocol=Vanilla, Cost ($)=0.652026.0678.43—
CAMELEvaluation Protocol=Vanilla, Cost ($)=0.232026.0669.93—
AutoRASEvaluation Protocol=Attack, Cost ($)=0.672026.06—82.35
CAMELEvaluation Protocol=Attack, Cost ($)=0.522026.06—62.75
Magentic-OneEvaluation Protocol=Attack, Cost ($)=0.382026.06—68.62
OWLEvaluation Protocol=Attack, Cost ($)=0.672026.06—68.63