List your company
ResearchFormerly WizWand

Stepwise tool-use on API-Bank (test)

74Success RateBoN+rubric (Ctrl)
67.7669.387172.62May 10, 2026
Updated 5mo ago

Evaluation results

MethodLinks
BoN+rubric (Ctrl)Model=o3-mini2026.0574
BoN+rubric (Ctrl)Model=GPT-4.12026.0574
Self-RefineModel=o3-mini2026.0573
Best-of-NModel=GPT-4.1-mini2026.0573
Best-of-NModel=GPT-4o2026.0573
Best-of-NModel=o3-mini2026.0573
Best-of-NModel=GPT-4.12026.0573
BoN+fixed rubric (Ctrl)Model=o3-mini2026.0573
BoN+rubric (Ctrl)Model=GPT-4.1-mini2026.0573
BoN+rubric (Ctrl)Model=GPT-4o2026.0573
CodeAct (Baseline)Model=GPT-4.1-mini2026.0572
CodeAct (Baseline)Model=GPT-4o2026.0572
CodeAct (Baseline)Model=o3-mini2026.0572
CodeAct (Baseline)Model=GPT-4.12026.0572
BoN+fixed rubric (Ctrl)Model=GPT-4.1-mini2026.0572
BoN+fixed rubric (Ctrl)Model=GPT-4o2026.0572
BoN+fixed rubric (Ctrl)Model=GPT-4.12026.0572
Fixed RubricRefine (Ctrl)Model=GPT-4.1-mini2026.0572
Fixed RubricRefine (Ctrl)Model=GPT-4o2026.0572
Fixed RubricRefine (Ctrl)Model=o3-mini2026.0572
Self-RefineModel=GPT-4o2026.0571
Self-RefineModel=GPT-4.12026.0571
Fixed RubricRefine (Ctrl)Model=GPT-4.12026.0571
RubricRefine (Ours)Model=GPT-4o2026.0570
RubricRefine (Ours)Model=GPT-4.1-mini2026.0569
RubricRefine (Ours)Model=GPT-4.12026.0569
Self-RefineModel=GPT-4.1-mini2026.0568
RubricRefine (Ours)Model=o3-mini2026.0568