Task-oriented Interaction on Meal Kit Recommendation Multi-turn
80Pass RateIn-context
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| In-contextBackbone=GPT-5 mini2026.03 | 80 | 158,552 | |
| Semantic XPathBackbone=Gemini 3 Flash, Scorer=Entail2026.03 | 80 | 6,782 | |
| Semantic XPathBackbone=GPT-5 mini, Scorer=Entail2026.03 | 70 | 6,079 | |
| Semantic XPathBackbone=Gemini 3 Flash, Scorer=Sem2026.03 | 70 | 6,954 | |
| Semantic XPathBackbone=GPT-5 mini, Scorer=Sem2026.03 | 60 | 7,087 | |
| In-contextBackbone=Gemini 3 Flash2026.03 | 40 | 174,567 | |
| Flat RAGBackbone=GPT-5 mini2026.03 | 10 | 3,686 | |
| Flat RAGBackbone=Gemini 3 Flash2026.03 | 10 | 3,596 |