Agentic Tool-Use on tau-bench Retail and Airline
78.3Retail Success RateModSFT+RL
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| ModSFT+RLBackbone=Qwen3-8B2026.06 | 78.3 | 16.1 | 47.2 | |
| Rejuvenation+RLBackbone=Qwen3-8B2026.06 | 77 | 17.5 | 47.3 | |
| DFT+RLBackbone=Qwen3-8B2026.06 | 75.3 | 12.2 | 43.8 | |
| OverSFT+RLBackbone=Qwen3-8B2026.06 | 70.3 | 19.9 | 45.1 |