Agent Performance on Tau-bench Telecom
45.39Avg@4 ScoreNoisyAgent
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| NoisyAgentbackbone=Qwen3-32B2026.05 | 45.39 | 78.07 | |
| GSPObackbone=Qwen3-32B2026.05 | 43.86 | 70.18 | |
| DAPObackbone=Qwen3-32B2026.05 | 43.42 | 71.93 | |
| GRPObackbone=Qwen3-32B2026.05 | 41.67 | 68.42 | |
| NoisyAgentbackbone=Qwen3-8B2026.05 | 40.79 | 64.91 | |
| DAPObackbone=Qwen3-8B2026.05 | 39.47 | 63.16 | |
| GSPObackbone=Qwen3-8B2026.05 | 39.04 | 61.4 | |
| GRPObackbone=Qwen3-8B2026.05 | 37.28 | 57.89 | |
| Qwen3-32Bbackbone=Qwen3-32B2026.05 | 28.95 | 49.12 | |
| Qwen3-8Bbackbone=Qwen3-8B2026.05 | 22.59 | 42.98 |