Web Agent Interaction on WebShop
88.4ScoreENVRL-GiGPO
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ENVRL-GiGPOBackbone Model=Qwen2.5-7B-Instruct2026.06 | 88.4 | 76.3 | |
| ENVRL-GiGPOBackbone Model=Qwen2.5-1.5B-Instruct2026.06 | 88.2 | 74.2 | |
| GiGPOBackbone Model=Qwen2.5-7B-Instruct2026.06 | 86.2 | 75.2 | |
| GiGPOBackbone Model=Qwen2.5-1.5B-Instruct2026.06 | 83.5 | 67.4 | |
| ENVRL-GRPOBackbone Model=Qwen2.5-1.5B-Instruct2026.06 | 83 | 67 | |
| ENVRL-GRPOBackbone Model=Qwen2.5-7B-Instruct2026.06 | 81.8 | 68.6 | |
| GRPOBackbone Model=Qwen2.5-7B-Instruct2026.06 | 79.3 | 66.1 | |
| GRPOBackbone Model=Qwen2.5-1.5B-Instruct2026.06 | 75.8 | 56.8 | |
| ReActBackbone Model=Qwen2.5-7B-Instruct2026.06 | 46.2 | 19.5 | |
| Gemini-2.5-ProBackbone Model=Proprietary Model2026.06 | 42.5 | 35.9 | |
| ReActBackbone Model=Qwen2.5-1.5B-Instruct2026.06 | 40.1 | 11.3 | |
| GPT-4oBackbone Model=Proprietary Model2026.06 | 31.8 | 23.7 | |
| PromptingBackbone Model=Qwen2.5-7B-Instruct2026.06 | 26.4 | 7.8 | |
| PromptingBackbone Model=Qwen2.5-1.5B-Instruct2026.06 | 23.1 | 5.2 |