Web interaction on Mind2Web (cross-website)
44.79Action AccuracyEDV
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| EDVBackbone Model=Ensemble2026.06 | 44.79 | 4.76 | 55.64 | 36.56 | |
| Reasoning Bank (RB)Backbone Model=GLM-4.7-FP8, Memory Strategy=RB2026.06 | 41.66 | 4.52 | 53.42 | 35.83 | |
| RouterBackbone Model=Ensemble2026.06 | 39.91 | 3.39 | 52.15 | 31.76 | |
| No Memory (NM)Backbone Model=GLM-4.7-FP8, Memory Strategy=NM2026.06 | 39.26 | 3.39 | 51.65 | 31.17 | |
| Reasoning Bank (RB)Backbone Model=Mimo-V2-Flash, Memory Strategy=RB2026.06 | 36.81 | 4.12 | 49.63 | 31.78 | |
| Reasoning Bank (RB)Backbone Model=Minimax-M2.1, Memory Strategy=RB2026.06 | 36.42 | 2.82 | 49.79 | 29.72 | |
| No Memory (NM)Backbone Model=Mimo-V2-Flash, Memory Strategy=NM2026.06 | 35.55 | 4.76 | 48.02 | 30.29 | |
| Evolving-RLSkill injection=false2026.05 | 35.15 | 3.39 | — | — | |
| Evolving-RLSkill injection=true2026.05 | 35.14 | 1.81 | — | — | |
| No Memory (NM)Backbone Model=Minimax-M2.1, Memory Strategy=NM2026.06 | 35.03 | 2.82 | 49.59 | 29.13 | |
| JudgeBackbone Model=Ensemble2026.06 | 35.01 | 1.87 | 46.85 | 27.63 | |
| GRPOSkill injection=true2026.05 | 26.2 | 0.9 | — | — | |
| GRPOSkill injection=false2026.05 | 24.61 | 0 | — | — | |
| ReasoningBank2026.05 | 17.76 | 0.56 | — | — | |
| Memento2026.05 | 15.46 | 0.56 | — | — | |
| Base ModelSkill injection=true2026.05 | 15.01 | 0 | — | — | |
| Base ModelSkill injection=false2026.05 | 7.95 | 0 | — | — |