Long Video Question Answering on VideoMME Long
69.7Accuracy (Long)GOPAgen
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| GOPAgenType=Agentic video understanding framework2026.06 | 69.7 | 66.7 | — | — | — | |
| Deep Video DiscoveryType=Agentic video understanding framework2026.06 | 68.6 | 67.3 | — | — | — | |
| Light-OmniBackbone=Qwen2.5-Omni-7B, Interactive Response Calls=1.0, Memorization Calls=153.52026.07 | 66.1 | — | 0.76 | 1.52 | 2.28 | |
| WorldMMBackbone=Qwen3-VL-8B, Interactive Response Calls=≤5.02026.07 | 66 | — | — | — | 20 | |
| M3-AgentBackbone=Qwen3-32B, Interactive Response Calls=2.5, Memorization Calls=81.22026.07 | 61.8 | — | 16.09 | 9.82 | 25.91 | |
| MR. VideoType=Agentic video understanding framework2026.06 | 61.6 | 61.8 | — | — | — | |
| GPT-4oAccess=Closed-Source2026.06 | 60.9 | 65.3 | — | — | — | |
| OpenAI o3Access=Closed-Source2026.06 | 60.6 | 63.2 | — | — | — | |
| Gemini-1.5-ProAccess=Closed-Source2026.06 | 58.6 | 67.4 | — | — | — | |
| Cloud-Only (CO)Deployment Strategy=Cloud-Only, Backbone Model=VideoLLaMA3-7B2026.06 | 46.67 | — | — | — | — | |
| Gemini-2.0-FlashAccess=Closed-Source2026.06 | 45.7 | 63 | — | — | — | |
| Co-VStreamDeployment Strategy=Edge-Cloud Collaboration2026.06 | 44.11 | — | — | — | — | |
| Edge-Cloud Hybrid (EC)Deployment Strategy=Edge-Cloud Hybrid, Restoration Model=AuraSR-v22026.06 | 43.33 | — | — | — | — | |
| Local-Only (LO)Deployment Strategy=Local-Only, Backbone Model=VideoLLaMA3-2B2026.06 | 34.11 | — | — | — | — | |
| AdaReTaKeAccess=Open-Source2026.06 | — | 65 | — | — | — | |
| InternVL2.5-78BAccess=Open-Source2026.06 | — | 62.6 | — | — | — | |
| mPLUG-Owl3Access=Open-Source2026.06 | — | 50.1 | — | — | — | |
| VideoLucyType=Agentic video understanding framework2026.06 | — | 66.8 | — | — | — | |
| VideoTreeType=Agentic video understanding framework2026.06 | — | 54.2 | — | — | — |