Customer Support Interaction on τ-Bench Retail Verified (test)
92Pass RateGPT-5.2
Evaluation Results
| Method | Links | |
|---|---|---|
| GPT-5.2User Simulator=Gemini-3-flash2026.05 | 92 | |
| Gemini-3-flashUser Simulator=Gemini-3-flash2026.05 | 88 | |
| Gemini-3-flashUser Simulator=GPT-5.22026.05 | 87 | |
| Gemini-2.5-flashUser Simulator=Gemini-3-flash2026.05 | 79 | |
| GPT-5.2User Simulator=GPT-5.22026.05 | 69 | |
| claude-sonnet-4.6User Simulator=Gemini-3-flash2026.05 | 67 | |
| deepseek-3.1User Simulator=GPT-5.22026.05 | 67 | |
| Gemini-2.5-flashUser Simulator=GPT-5.22026.05 | 60 | |
| Qwen-32BUser Simulator=GPT-5.22026.05 | 56 | |
| Qwen-32BUser Simulator=Gemini-3-flash2026.05 | 48 | |
| deepseek-3.1User Simulator=Gemini-3-flash2026.05 | 47 |