Multi-Turn Constraint Adaptation on FlexTravelBench 2-Turn Global Scenario
100Delivery RateReAct (GPT-4)
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| ReAct (GPT-4)Backbone=GPT-42026.03 | 100 | 61.93 | 63.75 | 51.67 | 68.34 | 26.25 | |
| ATLAS (Gemini-2.5-Pro)Backbone=Gemini-2.5-Pro2026.03 | 100 | 67.09 | 69.22 | 70.14 | 70 | 39.58 | |
| HiMAP-Travel (4B)Model Size=4B2026.03 | 100 | 68.2 | 70.15 | 71.45 | 71.2 | 40.85 | |
| HiMAP-Travel (8B)Model Size=8B2026.03 | 100 | 69.45 | 71.3 | 72.8 | 72.35 | 43.72 |