Web Navigation on WebShop unseen (test)
87.2ScoreProxMO
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ProxMOBackbone=Qwen2.5-7B-Instruct2026.02 | 87.2 | 76.5 | |
| GiGPOBackbone=Qwen2.5-7B-Instruct2026.02 | 85.5 | 74.8 | |
| ProxMOBackbone=Qwen2.5-1.5B-Instruct2026.02 | 85.3 | 67.1 | |
| GiGPOBackbone=Qwen2.5-1.5B-Instruct2026.02 | 81.7 | 62.3 | |
| GRPOBackbone=Qwen2.5-7B-Instruct2026.02 | 79.2 | 67.2 | |
| GRPOBackbone=Qwen2.5-1.5B-Instruct2026.02 | 73.1 | 52.2 | |
| ReflexionBackbone=Qwen2.5-1.5B-Instruct2026.02 | 58.6 | 23.5 | |
| ReflexionBackbone=Qwen2.5-7B-Instruct2026.02 | 56.3 | 30.2 | |
| ReActBackbone=Qwen2.5-7B-Instruct2026.02 | 47.8 | 21 | |
| Gemini-2.5-ProBackbone=Closed-Source LLMs2026.02 | 42.5 | 35.9 | |
| ReActBackbone=Qwen2.5-1.5B-Instruct2026.02 | 42.1 | 14.3 | |
| GPT-4oBackbone=Closed-Source LLMs2026.02 | 31.8 | 23.7 | |
| BaseBackbone=Qwen2.5-1.5B-Instruct2026.02 | 25.1 | 6.3 | |
| BaseBackbone=Qwen2.5-7B-Instruct2026.02 | 25.1 | 8.4 |