Web Agent Navigation on Mind2Web (Cross-Website)
46.1Execution Accuracy (EA)ReasoningBank
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| ReasoningBankBackbone=Gemini-2.5-pro2025.09 | 46.1 | 54.8 | 36.9 | 3.8 | |
| ReasoningBankBackbone=Gemini-2.5-flash2025.09 | 44.3 | 52.6 | 33.9 | 2.3 | |
| AWMBackbone=Gemini-2.5-pro2025.09 | 41.9 | 47.9 | 34.8 | 2.3 | |
| SynapseBackbone=Gemini-2.5-pro2025.09 | 41.8 | 51.2 | 35 | 3.2 | |
| No MemoryBackbone=Gemini-2.5-pro2025.09 | 41.2 | 49.8 | 34.8 | 3.4 | |
| SynapseBackbone=Gemini-2.5-flash2025.09 | 40.3 | 46 | 32.1 | 1.9 | |
| No MemoryBackbone=Gemini-2.5-flash2025.09 | 39.8 | 45.1 | 31.7 | 1.7 | |
| AWMBackbone=Gemini-2.5-flash2025.09 | 39.1 | 42.2 | 31.7 | 2.1 |