Computer-Using Agent Task on WebArena shopping_admin 12-task Warm (Run 2)
6.25Success RateMuscle-Mem
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Muscle-Memcache_type=blind linear cache, n_reps=4, backbone=Claude Sonnet 4.62026.06 | 6.25 | 0.75 | |
| PreAct gate-ON + cache-miss CUA fallbackn_reps=4, backbone=Claude Sonnet 4.62026.06 | 6.25 | 1 | |
| PreAct gate-ONverify_gate=enabled, n_reps=4, backbone=Claude Sonnet 4.62026.06 | 2 | 4 | |
| Workflow-Usescripts=per-task, n_reps=4, backbone=Claude Sonnet 4.62026.06 | 0 | 7 | |
| PreAct gate-OFFverify_gate=disabled, n_reps=4, backbone=Claude Sonnet 4.62026.06 | 0 | 5.75 |