Question Answering on WebQ (accuracy)
79.66Accuracy (WebQ)surprise
Evaluation results
| Method | Links | |
|---|---|---|
| surprise2026.06 | 79.66 | |
| TAA-k2026.06 | 79.66 | |
| Adaptive-k2026.06 | 78.18 | |
| Self-Route2026.06 | 76.91 | |
| ToolTreeBackbone Model=GPT-4o2026.03 | 67.94 | |
| ToolTreeBackbone Model=GPT-4o-mini2026.03 | 64.28 | |
| OctoToolsBackbone Model=GPT-4o2026.03 | 63.44 | |
| OctoToolsBackbone Model=GPT-4o-mini2026.03 | 61.12 | |
| HuggingGPTBackbone Model=GPT-4o2026.03 | 58.18 | |
| Few-ShotBackbone Model=GPT-4o2026.03 | 56.41 | |
| HuggingGPTBackbone Model=GPT-4o-mini2026.03 | 56.24 | |
| STITCH-RFull Duplex=false2026.07 | 50.3 | |
| Few-ShotBackbone Model=GPT-4o-mini2026.03 | 50.2 | |
| Qwen2-AudioFull Duplex=false2026.07 | 45.2 | |
| DuplexPOFull Duplex=true2026.07 | 44.5 | |
| SFT BaselineFull Duplex=true2026.07 | 44.3 | |
| SALMONN-omniFull Duplex=true2026.07 | 43.7 | |
| Baichuan-AudioFull Duplex=false2026.07 | 40.7 | |
| Kimi-AudioFull Duplex=false2026.07 | 37.3 | |
| GLM-4-VoiceFull Duplex=false2026.07 | 37 | |
| Freeze-OmniFull Duplex=true2026.07 | 27.9 | |
| SALM-DuplexFull Duplex=true2026.07 | 25 | |
| MoshiFull Duplex=true2026.07 | 22.1 |