Original Tasks on extended tau-Bench Retail domain
90Pass@1ReasoningBank
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| ReasoningBankModel=Claude-4.5-Opus2026.04 | 90 | 85 | 81 | 79 | |
| PolicyBankModel=Claude-4.5-Opus2026.04 | 89 | 84 | 82 | 80 | |
| ReasoningBankModel=Gemini-3-Pro2026.04 | 87 | 84 | 79 | 77 | |
| No memoryModel=Claude-4.5-Opus2026.04 | 87 | 82 | 78 | 75 | |
| AWMModel=Claude-4.5-Opus2026.04 | 87 | 84 | 80 | 78 | |
| AWMModel=Gemini-3-Pro2026.04 | 85 | 82 | 78 | 72 | |
| PolicyBankModel=Gemini-3-Pro2026.04 | 85 | 85 | 79 | 75 | |
| SynapseModel=Claude-4.5-Opus2026.04 | 85 | 80 | 80 | 77 | |
| No memoryModel=Gemini-3-Pro2026.04 | 82 | 77 | 73 | 70 | |
| SynapseModel=Gemini-3-Pro2026.04 | 81 | 80 | 75 | 70 |