Tool-use instruction following on ToolBench v1 (Evaluation)
71.7Success RateAgentMark
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| AgentMarkTask=T4: Multi-tool within-category selection2026.01 | 71.7 | 6.5 | 0.49 | 5.1 | -1.54 | 10.97 | |
| AgentMarkTask=T2: Single-tool instruction-following2026.01 | 61.7 | 8.6 | 0.48 | 4.89 | -1.41 | -16.96 | |
| AgentMarkTask=T1: Single-tool selection2026.01 | 60 | 5.6 | 0.51 | 5.28 | -1.67 | 9.64 | |
| AgentMarkTask=T3: Single-tool within-category selection2026.01 | 60 | 4.8 | 0.46 | 4.62 | -2.4 | -3.29 | |
| AgentMarkTask=Avg.2026.01 | 59.7 | 7.2 | 0.49 | 4.93 | -1.27 | -6.25 | |
| AgentMarkTask=T6: Complex multi-tool instruction-following2026.01 | 55 | 9.6 | 0.49 | 4.9 | 0.24 | -14.77 | |
| AgentMarkTask=T5: Multi-tool instruction-following2026.01 | 50 | 7.8 | 0.48 | 4.85 | -0.83 | 5.24 |