Agent task completion on Agent calling-tree and platform skill-graph testbeds
49.2Success Rate @ 1MPNN world model
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| MPNN world model2026.06 | 49.2 | 41.1 | 41.1 | 41.3 | |
| Error-Aware GWM2026.06 | 49.1 | 42 | 41.7 | 41.7 | |
| GCN world model2026.06 | 48.7 | 41.5 | 41.5 | 41.6 | |
| MLP world model2026.06 | 47.7 | 42 | 42.2 | 42.3 |