Interleaved Reactive-Proactive Requests on IPIBench
96R2P ScoreHuman Level
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Human Level2026.05 | 96 | 96 | 98 | 96.67 | |
| Qwen3.5-PlusModel Type=Open-source, Inference Mode=Offline2026.05 | 32.25 | 22.01 | 30.57 | 28.28 | |
| GPT-5.4Model Type=Proprietary, Inference Mode=Offline2026.05 | 27.04 | 15.06 | 29.28 | 23.79 | |
| GPT-4oModel Type=Proprietary, Inference Mode=Offline2026.05 | 27.04 | 17.68 | 26.17 | 23.63 | |
| Gemini 3 ProModel Type=Proprietary, Inference Mode=Offline2026.05 | 24.76 | 17.68 | 26.69 | 23.04 | |
| Qwen3-VL-8BModel Type=Open-source, Inference Mode=Offline2026.05 | 24.43 | 19.75 | 23.05 | 22.41 | |
| GLM-4.6VModel Type=Open-source, Inference Mode=Offline2026.05 | 22.15 | 6.77 | 39.37 | 22.76 | |
| Gemini 2.5 ProModel Type=Proprietary, Inference Mode=Offline2026.05 | 21.17 | 17.54 | 28.24 | 22.32 | |
| InternVL3-8BModel Type=Open-source, Inference Mode=Offline2026.05 | 7.17 | 0 | 9.33 | 5.5 | |
| LLaVA-OneVision-7BModel Type=Open-source, Inference Mode=Offline2026.05 | 3.91 | 0 | 3.89 | 2.6 | |
| VideoLLM-online-8BModel Type=Open-source, Inference Mode=Online2026.05 | 0 | 1.1 | 1.55 | 0.88 | |
| Flash-VStream-7BModel Type=Open-source, Inference Mode=Online2026.05 | 0 | 0 | 0.78 | 0.26 |