Multi-turn agent interaction on WebShop
89.06Success RateATOD
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| ATODStudent Model=Qwen3-1.7B, Teacher Model=Qwen3-4B GRPO2026.06 | 89.06 | — | 6.2 | |
| ATODStudent Model=Qwen3-0.6B, Teacher Model=Qwen3-4B GRPO2026.06 | 86.72 | — | 7.48 | |
| OPDStudent Model=Qwen3-0.6B, Teacher Model=Qwen3-4B GRPO2026.06 | 85.16 | — | 6.23 | |
| TCODStudent Model=Qwen3-0.6B, Teacher Model=Qwen3-4B GRPO2026.06 | 84.38 | — | 5.94 | |
| GRPOStudent Model=None, Teacher Model=Qwen3-4B GRPO2026.06 | 82.03 | — | 6.21 | |
| GRPOStudent Model=Qwen3-4B, Teacher Model=None2026.06 | 82.03 | — | 6.21 | |
| TCODStudent Model=Qwen3-1.7B, Teacher Model=Qwen3-4B GRPO2026.06 | 81.25 | — | 6.1 | |
| SODStudent Model=Qwen3-0.6B, Teacher Model=Qwen3-4B GRPO2026.06 | 80.47 | — | 5.95 | |
| ATODStudent Model=Qwen3-4B, Teacher Model=Qwen3-30B-A3B GRPO2026.06 | 78.91 | — | 8.29 | |
| OPDStudent Model=Qwen3-1.7B, Teacher Model=Qwen3-4B GRPO2026.06 | 77.34 | — | 6.15 | |
| SODStudent Model=Qwen3-1.7B, Teacher Model=Qwen3-4B GRPO2026.06 | 76.56 | — | 5.59 | |
| SODStudent Model=Qwen3-4B, Teacher Model=Qwen3-30B-A3B GRPO2026.06 | 76.56 | — | 6.68 | |
| GRPOStudent Model=None, Teacher Model=Qwen3-30B-A3B GRPO2026.06 | 75.78 | — | 8.02 | |
| TCODStudent Model=Qwen3-4B, Teacher Model=Qwen3-30B-A3B GRPO2026.06 | 75.78 | — | 9.21 | |
| OPDStudent Model=Qwen3-4B, Teacher Model=Qwen3-30B-A3B GRPO2026.06 | 75 | — | 8.53 | |
| SDARStudent Model=Qwen3-4B, Teacher Model=None2026.06 | 67.97 | — | 7.91 | |
| GRPOStudent Model=Qwen3-1.7B, Teacher Model=None2026.06 | 47.66 | — | 10.01 | |
| SDARStudent Model=Qwen3-1.7B, Teacher Model=None2026.06 | 43.75 | — | 8.9 | |
| GRPOStudent Model=Qwen3-0.6B, Teacher Model=None2026.06 | 29.69 | — | 9.73 | |
| Zero-shotParams=30B-A3B2026.06 | 19.25 | 55.07 | 4.96 | |
| TCODParams=4B2026.06 | 19 | 53.13 | 4.26 | |
| Guided-OPDParams=4B2026.06 | 17.75 | 55.78 | 4.28 | |
| TCODParams=1.7B2026.06 | 14.25 | 49.47 | 4.39 | |
| Vanilla OPDParams=4B2026.06 | 13.5 | 50.87 | 4.53 | |
| AdaSwitchParams=4B2026.06 | 12.35 | 48.91 | 5.57 | |
| Guided-OPDParams=1.7B2026.06 | 11.75 | 52.63 | 4.14 | |
| Vanilla OPDParams=1.7B2026.06 | 11 | 47.59 | 5.54 | |
| AdaSwitchParams=1.7B2026.06 | 10.5 | 46.79 | 5.81 | |
| Zero-shotParams=1.7B2026.06 | 10.25 | 28.71 | 9.11 | |
| Guided-OPDParams=0.6B2026.06 | 9.75 | 48.44 | 4.8 | |
| SDARStudent Model=Qwen3-0.6B, Teacher Model=None2026.06 | 8.59 | — | 10.27 | |
| TCODParams=0.6B2026.06 | 8.25 | 46.37 | 3.83 | |
| Zero-shotParams=4B2026.06 | 8.25 | 20.59 | 12.23 | |
| Vanilla OPDParams=0.6B2026.06 | 8 | 39.89 | 6.95 | |
| AdaSwitchParams=0.6B2026.06 | 7.75 | 43.41 | 5.08 | |
| VanillaStudent Model=Qwen3-1.7B, Teacher Model=None2026.06 | 5.47 | — | 8.43 | |
| VanillaStudent Model=Qwen3-0.6B, Teacher Model=None2026.06 | 4.69 | — | 9.59 | |
| Zero-shotParams=0.6B2026.06 | 3.75 | 10.37 | 14.03 | |
| VanillaStudent Model=Qwen3-4B, Teacher Model=None2026.06 | 0.78 | — | 13.99 |