Task-oriented Interaction on Travel Itinerary Multi-turn
90Pass RateIn-context
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| In-contextBackbone=GPT-5 mini2026.03 | 90 | 28,342 | |
| Semantic XPathBackbone=GPT-5 mini, Scorer=Entail2026.03 | 80 | 5,820 | |
| In-contextBackbone=Gemini 3 Flash2026.03 | 60 | 31,639 | |
| Semantic XPathBackbone=Gemini 3 Flash, Scorer=Sem2026.03 | 60 | 5,451 | |
| Semantic XPathBackbone=Gemini 3 Flash, Scorer=Entail2026.03 | 60 | 5,671 | |
| Semantic XPathBackbone=GPT-5 mini, Scorer=Sem2026.03 | 50 | 5,664 | |
| Flat RAGBackbone=GPT-5 mini2026.03 | 20 | 3,250 | |
| Flat RAGBackbone=Gemini 3 Flash2026.03 | 20 | 3,164 |