Tool Calling on WHEN2TOOL Overall
-1Δ AccuracyNecessary (N)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Necessary (N)model averaging=six models2026.05 | -1 | -0.06 | -16.8 | |
| PROBE&PREFILLmodel averaging=six models, threshold (τ)=0.52026.05 | -1.7 | -0.48 | -3.6 | |
| Sparse (S)model averaging=six models2026.05 | -8.4 | -0.46 | -18.4 | |
| Necessary + Reason-then-Actmodel averaging=six models2026.05 | -15.8 | -0.82 | -19.2 | |
| Sparse + Reason-then-Actmodel averaging=six models2026.05 | -19.7 | -1 | -19.6 | |
| No Tool + Reason-then-Actmodel averaging=six models2026.05 | -24.2 | -1.08 | -22.4 | |
| No Tool (X)model averaging=six models2026.05 | -28.7 | -0.71 | -40.5 |