Implicit-need coverage on 100-query four-scene benchmark v2 (full)
80.4Overall CoverageAURA Intent
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| AURA IntentBackbone=gpt-4o-mini, Temperature=0.1, Seeds=3, Probe ceiling=0–32026.06 | 80.4 | 80 | 77.9 | 81.1 | 82.7 | |
| NoIntentBackbone=gpt-4o-mini, Temperature=0.1, Seeds=3, Style=ReAct, Budget=32026.06 | 73.3 | 70.9 | 68.8 | 73.1 | 80.3 | |
| LiteralBackbone=gpt-4o-mini, Temperature=0.1, Seeds=3, Tools=None2026.06 | 22.3 | 20 | 26.4 | 23.7 | 19.2 |