Agentic Tool-use on τ²-bench airline v1 (test)
53.5Pass@1RL
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| RLBackbone=Qwen3-8B, Protocol=Single-Task2026.06 | 53.5 | 33.6 | |
| On-policy Distill.Backbone=Qwen3-8B, Protocol=Multi-Task2026.06 | 52.6 | 32.9 | |
| Off-policy Distill.Backbone=Qwen3-8B, Protocol=Multi-Task2026.06 | 51.8 | 31.3 | |
| Off. + On.Backbone=Qwen3-8B, Protocol=Multi-Task2026.06 | 51.4 | 33.2 | |
| Off-policy Distill.Backbone=Qwen3-8B, Protocol=Single-Task2026.06 | 50 | 32.1 | |
| Pre-RLBackbone=Qwen3-8B, Protocol=Single-Task2026.06 | 42.5 | 26.5 |