Multi-turn Agentic Tool-use on WebShop
79.6Success RateHERO
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| HEROModel=Qwen3-30B-A3B-Instruct2026.06 | 79.6 | 7 | |
| GRPOModel=Qwen3-30B-A3B-Instruct2026.06 | 76.1 | 7.3 | |
| HEROModel=Qwen3-4B-Instruct2026.06 | 68.9 | 8.1 | |
| GRPOModel=Qwen3-4B-Instruct2026.06 | 65.2 | 8.1 | |
| Environment Feedback OnlyModel=Qwen3-30B-A3B-Instruct2026.06 | 44.3 | 9.3 | |
| Environment Feedback OnlyModel=Qwen3-4B-Instruct2026.06 | 37.5 | 10.4 | |
| BaseModel=Qwen3-30B-A3B-Instruct2026.06 | 32.7 | 8.5 | |
| Full-Demo Privileged TeacherModel=Qwen3-30B-A3B-Instruct2026.06 | 25.9 | 9.9 | |
| BaseModel=Qwen3-4B-Instruct2026.06 | 21.8 | 8.6 | |
| Full-Demo Privileged TeacherModel=Qwen3-4B-Instruct2026.06 | 17.4 | 10.6 |