Multi-turn Agentic Tool-use on TauBench-Airline (OOD)
35.5Success Rate (SR)HERO
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| HEROModel=Qwen3-30B-A3B-Instruct2026.06 | 35.5 | 11.2 | |
| GRPOModel=Qwen3-30B-A3B-Instruct2026.06 | 33.5 | 11.6 | |
| Environment Feedback OnlyModel=Qwen3-30B-A3B-Instruct2026.06 | 32 | 12.6 | |
| BaseModel=Qwen3-30B-A3B-Instruct2026.06 | 30 | 12.3 | |
| Full-Demo Privileged TeacherModel=Qwen3-30B-A3B-Instruct2026.06 | 26 | 13.4 | |
| HEROModel=Qwen3-4B-Instruct2026.06 | 19.5 | 10.1 | |
| GRPOModel=Qwen3-4B-Instruct2026.06 | 18 | 14 | |
| Environment Feedback OnlyModel=Qwen3-4B-Instruct2026.06 | 12.5 | 13.7 | |
| BaseModel=Qwen3-4B-Instruct2026.06 | 12 | 13.1 | |
| Full-Demo Privileged TeacherModel=Qwen3-4B-Instruct2026.06 | 8.5 | 13.9 |