Long-Horizon Agent Tasks on AppWorld (test)
86.2Easy Success Rate (SR)OCR-Memory
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| OCR-Memory2026.04 | 86.2 | 57.4 | 30.8 | 58.1 | |
| ACON2026.04 | 84.8 | 55.1 | 28.7 | 56.2 | |
| AWM2026.04 | 84.1 | 53.6 | 27.2 | 55 | |
| MemoryBank2026.04 | 81.3 | 50.1 | 24.9 | 52.1 | |
| Retrievalmode=Retrieval-based2026.04 | 72.5 | 44.8 | 21.4 | 46.2 | |
| Zero-Shotmode=Zero-Shot2026.04 | 68.7 | 36.2 | 20.9 | 41.9 |