Task Goal Completion on AppWorld Average 168 tasks (test normal)
44AccuracyNo compression
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| No compression2026.06 | 44 | 19.1 | 92.3 | 0 | |
| LRE2026.06 | 41.1 | 21.8 | 69.1 | 0 | |
| FIFOstrategy=recency2026.06 | 34.5 | 30.6 | 56.5 | 0 | |
| ACONmode=in-fork2026.06 | 26.2 | 33.5 | 51.2 | 5.45 |