Tool-use Task Completion on Tau2-Bench Telecom subset Solo mode
43.3Success RateReAct + EvoSOP
Evaluation Results
| Method | Links | |
|---|---|---|
| ReAct + EvoSOPBackbone=Gemini-3-FP2026.07 | 43.3 | |
| ReAct + EvoSOPBackbone=GPT-4o2026.07 | 40.9 | |
| ReAct + EvoSOPBackbone=Qwen-Max2026.07 | 40.2 | |
| ReAct + ASIBackbone=Qwen-Max2026.07 | 39.5 | |
| ReAct + ASIBackbone=GPT-4o2026.07 | 37.7 | |
| ReActBackbone=GPT-4o2026.07 | 37.1 | |
| ReAct + ASIBackbone=Gemini-3-FP2026.07 | 35.4 |