Tool Orchestration on StableToolBench 1.0 (test)
50.3I1 Instruction Success RateRETO
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| RETOBackbone=Qwen2.5-7B, Tool-tuned=false2026.02 | 50.3 | 50.9 | 54.4 | 46.5 | 52.4 | 42.5 | 49.5 | |
| ReActBackbone=GPT-3.5-0613, Tool-tuned=N/A2026.02 | 50.1 | 51.6 | 49.8 | 37.3 | 40.3 | 43.2 | 45.4 | |
| ReActBackbone=GPT-3.5-1106, Tool-tuned=N/A2026.02 | 49.8 | 51.3 | 40.1 | 39.8 | 49.2 | 59.6 | 48.3 | |
| DFSDTBackbone=ToolLLaMA-7B, Tool-tuned=true2026.02 | 48.8 | 48.9 | 48.9 | 37.3 | 45.8 | 53 | 47.1 | |
| ReActBackbone=ToolLLaMA-7B, Tool-tuned=true2026.02 | 31.5 | 34 | 39.5 | 25.5 | 34.5 | 32.8 | 33 | |
| RETOBackbone=LLaMA-3.1-8B, Tool-tuned=false2026.02 | 14 | 17.4 | 22.9 | 16.7 | 16.8 | 14.5 | 17.1 | |
| ReActBackbone=Qwen2.5-7B, Tool-tuned=false2026.02 | 8.4 | 11 | 8.1 | 11.3 | 17.2 | 1.1 | 9.5 | |
| DFSDTBackbone=Qwen2.5-7B, Tool-tuned=false2026.02 | 5.7 | 15.3 | 10.1 | 17.1 | 24.7 | 0.5 | 12.2 | |
| ReActBackbone=LLaMA-3.1-8B, Tool-tuned=false2026.02 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | |
| DFSDTBackbone=LLaMA-3.1-8B, Tool-tuned=false2026.02 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |