Tool-use agent evaluation on τ-bench retail
44.173Pass@1 Success RateFAMA
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| FAMABackbone=Qwen2.5-72B-Instruct2026.04 | 44.173 | 34.26 | 30.26 | 28.17 | 26.95 | |
| ReActBackbone=Qwen2.5-72B-Instruct2026.04 | 43.47 | 32.26 | 26.69 | 23.13 | 20.86 | |
| SRBackbone=Qwen2.5-72B-Instruct2026.04 | 42.9 | 30.7 | 25 | 21.2 | 18.2 | |
| FAMABackbone=Qwen3-32B2026.04 | 40.5 | 26.9 | 19.9 | 15.3 | 12.2 | |
| ReActBackbone=Qwen3-32B2026.04 | 38 | 25 | 18.8 | 15.3 | 10 | |
| SRBackbone=Qwen3-32B2026.04 | 26.7 | 15.6 | 11.3 | 9.2 | 7.8 |