Conversational Agent Interaction on Du-Interact Golden Benchmark
82.5Dispatch Rate (P_disp)DuCCAE-V3
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| DuCCAE-V3Setup=Evolution II2026.02 | 82.5 | 72.4 | 71.1 | 4.1 | 4.3 | 1,880 | |
| DuCCAE-V2Setup=Evolution I2026.02 | 76.2 | 63.8 | 66.5 | 3.9 | 4 | 1,920 | |
| Llama-3.3-70B-InstructModel Tier=Large Scale Class (30B-70B), Setup=Zero-shot2026.02 | 75.5 | 61.1 | 64.4 | 3.4 | 3.3 | 5,800 | |
| Qwen2.5-32B-InstructModel Tier=Large Scale Class (30B-70B), Setup=Zero-shot2026.02 | 72.1 | 55.4 | 62.2 | 3.2 | 3.2 | 3,250 | |
| DuCCAE-V1Setup=Cold Start2026.02 | 68.8 | 50.5 | 58.4 | 3.5 | 3.5 | 1,850 | |
| Qwen2.5-7B-InstructModel Tier=Server Class (7B-11B), Setup=Zero-shot2026.02 | 61.5 | 37.2 | 54.8 | 2.8 | 2.8 | 1,480 | |
| Llama-3.2-11B-InstructModel Tier=Server Class (7B-11B), Setup=Zero-shot2026.02 | 58.2 | 35.8 | 53.5 | 2.7 | 2.7 | 1,520 | |
| Qwen2.5-3B-InstructModel Tier=Edge Class (3B), Setup=Zero-shot2026.02 | 52.4 | 28.5 | 45.2 | 2.5 | 2.4 | 1,250 | |
| Llama-3.2-3B-InstructModel Tier=Edge Class (3B), Setup=Zero-shot2026.02 | 51.8 | 26.2 | 44.1 | 2.4 | 2.3 | 1,280 |