Tool Use on ToolBench Hard v1.0
97Tool Call Rate (TCR)NaviAgent
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| NaviAgentModel=Deepseek-V32025.06 | 97 | 44.9 | 5.19 | |
| NaviAgentModel=Qwen2.5-32B2025.06 | 79.4 | 30.8 | 5.31 | |
| α-UMIModel=Qwen2.5-32B2025.06 | 77.1 | 22.9 | 6.58 | |
| α-UMIModel=Deepseek-V32025.06 | 73 | 29.5 | 6.64 | |
| α-UMIModel=Qwen2.5-14B2025.06 | 67.6 | 13.3 | 6.07 | |
| NaviAgentModel=Qwen2.5-14B2025.06 | 61.1 | 22.4 | 4.68 | |
| ToolLLMModel=Qwen2.5-32B2025.06 | 49.5 | 23.8 | 4.1 | |
| ReActModel=Deepseek-V32025.06 | 48.6 | 23.8 | 3.74 | |
| ToolLLMModel=Qwen2.5-14B2025.06 | 38.1 | 11.4 | 4.06 | |
| ReActModel=Qwen2.5-32B2025.06 | 30.5 | 19 | 3.95 | |
| ToolLLMModel=Deepseek-V32025.06 | 29.7 | 24.8 | 3.9 | |
| ReActModel=Qwen2.5-14B2025.06 | 24.8 | 20 | 3.64 |