Agentic Tool-Use on Tau-Bench
71.3Retail ScoreQwen3-235B-Instruct-2507
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Qwen3-235B-Instruct-2507Mode=Instruct2026.01 | 71.3 | 44 | 57.65 | |
| Kimi-K2-Instruct-09052026.01 | 69.87 | 54 | 61.94 | |
| Qwen3-235B-Thinking-2507Mode=Thinking2026.01 | 67.8 | 46 | 56.9 | |
| GPT-4.12026.01 | 66.95 | 48 | 57.48 | |
| EnvScalerBackbone=Qwen3-8B, Training=SFT & RL2026.01 | 53.62 | 36 | 44.81 | |
| EnvScalerBackbone=Qwen3-8B, Training=SFT2026.01 | 48.7 | 34 | 41.35 | |
| EnvScalerBackbone=Qwen3-4B, Training=SFT & RL2026.01 | 48.12 | 34 | 41.06 | |
| Qwen3-8BMode=Thinking2026.01 | 46.38 | 30 | 38.19 | |
| EnvScalerBackbone=Qwen3-4B, Training=SFT2026.01 | 44.35 | 32 | 38.2 | |
| Qwen3-4BMode=Thinking2026.01 | 40.87 | 26 | 33.44 | |
| EnvScalerBackbone=Qwen3-1.7B, Training=SFT2026.01 | 20.87 | 14 | 17.44 | |
| EnvScalerBackbone=Qwen3-1.7B, Training=SFT & RL2026.01 | 18.55 | 14 | 16.28 | |
| Qwen3-1.7BMode=Thinking2026.01 | 8.99 | 16 | 12.5 |