Multi-Turn Constraint Adaptation on FlexTravelBench 2-Turn Local Scenario
100Delivery Success RateReAct (GPT-4)
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| ReAct (GPT-4)Backbone=GPT-42026.03 | 100 | 85.84 | 44.41 | 67.29 | 43.72 | 29.58 | |
| ATLAS (Gemini-2.5-Pro)Backbone=Gemini-2.5-Pro2026.03 | 100 | 75.48 | 59.8 | 77.23 | 64.62 | 38.86 | |
| HiMAP-Travel (4B)Model Size=4B2026.03 | 100 | 76.8 | 61.25 | 79.15 | 66.5 | 41.25 | |
| HiMAP-Travel (8B)Model Size=8B2026.03 | 100 | 78.62 | 62.15 | 80.35 | 67.4 | 44.96 |