Agent Performance on Tau-bench Retail
60.31Avg@4NoisyAgent
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| NoisyAgentbackbone=Qwen3-32B2026.05 | 60.31 | 86.84 | |
| GSPObackbone=Qwen3-32B2026.05 | 58.55 | 84.21 | |
| GRPObackbone=Qwen3-32B2026.05 | 58.11 | 83.33 | |
| DAPObackbone=Qwen3-32B2026.05 | 56.58 | 80.7 | |
| Qwen3-32Bbackbone=Qwen3-32B2026.05 | 49.12 | 72.81 | |
| NoisyAgentbackbone=Qwen3-8B2026.05 | 47.59 | 77.19 | |
| GSPObackbone=Qwen3-8B2026.05 | 46.49 | 74.56 | |
| GRPObackbone=Qwen3-8B2026.05 | 46.05 | 73.68 | |
| DAPObackbone=Qwen3-8B2026.05 | 44.52 | 71.05 | |
| Qwen3-8Bbackbone=Qwen3-8B2026.05 | 35.31 | 59.65 |