Long-Horizon Agent Tasks on Mind2Web (test)
53.8Element AccuracyOCR-Memory
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| OCR-Memory2026.04 | 53.8 | 59.2 | 46.1 | 4.8 | |
| AWM2026.04 | 49.1 | 55.7 | 42.6 | 4.3 | |
| ACON2026.04 | 48.2 | 54.1 | 41.4 | 4.1 | |
| MemoryBank2026.04 | 43.8 | 49.5 | 39.2 | 3.3 | |
| Retrievalmode=Retrieval-based2026.04 | 41.3 | 48.2 | 38.9 | 2.7 | |
| Zero-Shotmode=Zero-Shot2026.04 | 40.1 | 46.2 | 37.9 | 2.2 |