Tool-use Task Completion on ACEBench agent subset Multi-Turn
84.4Success RateReAct + EvoSOP
Evaluation Results
| Method | Links | |
|---|---|---|
| ReAct + EvoSOPBackbone=Qwen-Max2026.07 | 84.4 | |
| DFSDT + EvoSOPBackbone=Gemini-3-FP2026.07 | 83.3 | |
| ReAct + EvoSOPBackbone=Gemini-3-FP2026.07 | 79.4 | |
| ReAct + EvoSOPBackbone=GPT-4o2026.07 | 79.4 | |
| DFSDT + DRAFTBackbone=Qwen-Max2026.07 | 78.3 | |
| DFSDT + EvoSOPBackbone=Qwen-Max2026.07 | 77.9 | |
| DFSDT + EvoSOPBackbone=GPT-4o2026.07 | 74.4 | |
| ReActBackbone=GPT-4o2026.07 | 72.2 | |
| DFSDT + DRAFTBackbone=Gemini-3-FP2026.07 | 71.1 | |
| DFSDTBackbone=GPT-4o2026.07 | 69.4 | |
| DFSDT + DRAFTBackbone=GPT-4o2026.07 | 66.1 | |
| ReAct + ASIBackbone=Qwen-Max2026.07 | 65.6 | |
| ReAct + ASIBackbone=GPT-4o2026.07 | 57.2 | |
| ReAct + ASIBackbone=Gemini-3-FP2026.07 | 46.7 |