Video Understanding on MVBench (test)
100.3AccuracyFSR
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| FSRReduction Ratio=50%, Frames=322026.02 | 100.3 | 59.7 | |
| LLaVA-Video-7B-qwen2Reduction Ratio=100%, Frames=322026.02 | 100 | 60.8 | |
| FSRReduction Ratio=60%, Frames=322026.02 | 99.6 | 59.4 | |
| HoloVReduction Ratio=50%, Frames=322026.02 | 99.2 | 59.4 | |
| FSRReduction Ratio=70%, Frames=322026.02 | 98.9 | 59.2 | |
| HoloVReduction Ratio=60%, Frames=322026.02 | 98.5 | 59.3 | |
| HoloVReduction Ratio=70%, Frames=322026.02 | 98.2 | 59 | |
| FSRReduction Ratio=80%, Frames=322026.02 | 98.2 | 58.5 | |
| HoloVReduction Ratio=80%, Frames=322026.02 | 98 | 58.3 | |
| PLM-8BAccess Type=Open models2026.01 | 77.1 | — | |
| PLM-8BCategory=Fully open2026.03 | 77.1 | — | |
| Molmo2-8BAccess Type=Molmo2 family2026.01 | 75.9 | — | |
| Molmo2-8BCategory=Fully open2026.03 | 75.9 | — | |
| MolmoPoint-8BCategory=MolmoPoint2026.03 | 75.2 | — | |
| Molmo2-4BAccess Type=Molmo2 family2026.01 | 75.1 | — | |
| Molmo2-4BCategory=Fully open2026.03 | 75.1 | — | |
| Eagle2.5-8BAccess Type=Open weights only2026.01 | 74.8 | — | |
| Molmo2-O-7BAccess Type=Molmo2 family2026.01 | 74.8 | — | |
| Eagle2.5-8BCategory=Open weights2026.03 | 74.8 | — | |
| Molmo2-O-7BCategory=Fully open2026.03 | 74.8 | — | |
| PLM-3BAccess Type=Open models2026.01 | 74.7 | — | |
| PLM-3BCategory=Fully open2026.03 | 74.7 | — | |
| GPT-5Access Type=API call only2026.01 | 74.1 | — | |
| GPT-5type=Proprietary2026.02 | 74.1 | — | |
| GPT-5Category=API call only2026.03 | 74.1 | — | |
| VideoChat-Flash-7BAccess Type=Open models2026.01 | 74 | — | |
| VideoChat-Flash-7BCategory=Fully open2026.03 | 74 | — | |
| Qwen2-VL-72BParameters=72B2024.09 | 73.6 | — | |
| InternVL3.5-8BAccess Type=Open weights only2026.01 | 72.1 | — | |
| InternVL3.5-8BCategory=Open weights2026.03 | 72.1 | — | |
| InternVL2.5Model Size=7-8B, Zero-shot evaluation=true2025.12 | 72 | — | |
| InternVL3.5-4BAccess Type=Open weights only2026.01 | 71.2 | — | |
| InternVL3.5-4BCategory=Open weights2026.03 | 71.2 | — | |
| Gemini 2.5 ProAccess Type=API call only2026.01 | 70.6 | — | |
| Gemini 2.5 Protype=Proprietary2026.02 | 70.6 | — | |
| Gemini 2.5 ProCategory=API call only2026.03 | 70.6 | — | |
| Gemini 3 ProAccess Type=API call only2026.01 | 70.4 | — | |
| Gemini 3 Protype=Proprietary2026.02 | 70.4 | — | |
| Gemini 3 ProCategory=API call only2026.03 | 70.4 | — | |
| Oryx-1.5Size=32B2024.09 | 70.1 | — | |
| Structured Reward Reinforcement LearningBackbone Model=Qwen3-VL-32B, Method/Configuration=+Ours, Model Access Level=Open-source2026.04 | 69.8 | — | |
| Previous SoTA2024.09 | 69.6 | — | |
| Qwen2.5-VLModel Size=7-8B, Zero-shot evaluation=true2025.12 | 69.6 | — | |
| SeasonBackbone Model=Qwen3-VL-32B, Method/Configuration=+Season, Model Access Level=Open-source2026.04 | 69.6 | — | |
| Qwen3-VL-32BBackbone Model=Qwen3-VL-32B, Method/Configuration=Base, Model Access Level=Open-source2026.04 | 69.5 | — | |
| IXC-2.5-7Bnumber of parameters=7B2026.02 | 69.1 | — | |
| IXC-2.5-7BBackbone=7B2026.05 | 69.1 | — | |
| IXC-2.5# Params=7B2026.06 | 69.1 | — | |
| Qwen3-VL-4BAccess Type=Open weights only2026.01 | 68.9 | — | |
| Qwen3-VL-4BCategory=Open weights2026.03 | 68.9 | — | |
| Qwen3-VL-8BAccess Type=Open weights only2026.01 | 68.7 | — | |
| Qwen3-VL-8BCategory=Open weights2026.03 | 68.7 | — | |
| GLM-4.1V-9BAccess Type=Open weights only2026.01 | 68.4 | — | |
| JavisGPTModel Size=7-8B, #Samples=1.5M, Zero-shot evaluation=true2025.12 | 68.4 | — | |
| GLM-4.1V-9BCategory=Open weights2026.03 | 68.4 | — | |
| NVILANumber of Parameters=8B, #Frames=2562024.12 | 68.1 | — | |
| VIDEOP2RBase Model=Qwen2.5-VL-7B2025.11 | 68.1 | — | |
| NVILA-8BBackbone=8B2026.05 | 68.1 | — | |
| VideoChat-R1Base Model=Qwen2.5-VL-7B2025.11 | 67.9 | — | |
| Oryx-1.5Size=7B2024.09 | 67.6 | — | |
| Oryx-1.5Number of Parameters=8B, #Frames=1282024.12 | 67.6 | — | |
| Oryx-1.5-7BBackbone=1.5B2026.05 | 67.6 | — | |
| InternVideo2-7BBackbone=7B2026.05 | 67.2 | — | |
| LongVILALLM Size=7B2024.08 | 67.1 | — | |
| LongVILANumber of Parameters=7B, #Frames=2562024.12 | 67.1 | — | |
| Qwen2-VL-7BParameters=7B2024.09 | 67 | — | |
| Qwen2-VLNumber of Parameters=8B, #Frames=2fps2024.12 | 67 | — | |
| Qwen2-VL-7B-Ins.Backbone=7B-Instruct2024.12 | 67 | — | |
| Gemini 2.5 FlashAccess Type=API call only2026.01 | 67 | — | |
| Qwen2-VLModel Size=7-8B, Zero-shot evaluation=true2025.12 | 67 | — | |
| Gemini 2.5 FlashCategory=API call only2026.03 | 67 | — | |
| LongVUNumber of Parameters=7B, #Frames=1fps2024.12 | 66.9 | — | |
| Qwen2.5-OmniModel Size=7-8B, Zero-shot evaluation=true2025.12 | 66.9 | — | |
| Structured Reward Reinforcement LearningBackbone Model=Qwen3-VL-8B, Method/Configuration=+Ours, Model Access Level=Open-source2026.04 | 66.7 | — | |
| GPT-5 miniAccess Type=API call only2026.01 | 66.5 | — | |
| GPT-5 miniCategory=API call only2026.03 | 66.5 | — | |
| InternVL-2-8BBackbone=8B2024.12 | 66.4 | — | |
| Video-ToCFrames=642026.04 | 66.4 | — | |
| Video-ToCFrames=322026.04 | 66.3 | — | |
| InternVL2Number of Parameters=8B, #Frames=642024.12 | 65.8 | — | |
| InternVL2-8Bnumber of parameters=8B2026.02 | 65.8 | — | |
| Qwen3-VL-8BBackbone Model=Qwen3-VL-8B, Method/Configuration=Base, Model Access Level=Open-source2026.04 | 65.8 | — | |
| SeasonBackbone Model=Qwen3-VL-8B, Method/Configuration=+Season, Model Access Level=Open-source2026.04 | 65.7 | — | |
| Video-ToC-SFTFrames=642026.04 | 65.7 | — | |
| Video-ToC-SFTFrames=322026.04 | 65.3 | — | |
| Video-ToCFrames=162026.04 | 65 | — | |
| Video-R1Frames=642026.04 | 64.8 | — | |
| OryxSize=34B2024.09 | 64.7 | — | |
| VideoCAMSize=9B2024.09 | 64.6 | — | |
| VideoCCAM-9Bnumber of parameters=9B2026.02 | 64.6 | — | |
| SeasonBackbone Model=Qwen3-VL-4B, Method/Configuration=+Season, Model Access Level=Open-source2026.04 | 64.3 | — | |
| Structured Reward Reinforcement LearningBackbone Model=Qwen3-VL-4B, Method/Configuration=+Ours, Model Access Level=Open-source2026.04 | 64.2 | — | |
| VideoChat-R1Frames=162026.04 | 64.2 | — | |
| OryxSize=7B2024.09 | 63.9 | — | |
| Video-R1Base Model=Qwen2.5-VL-7B2025.11 | 63.9 | — | |
| Video-R1Frames=322026.04 | 63.9 | — | |
| Video-ToC-SFTFrames=162026.04 | 63.3 | — | |
| Qwen2-VL-2BParameters=2B2024.09 | 63.2 | — | |
| Time-R1Base Model=Qwen2.5-VL-7B2025.11 | 63.1 | — | |
| VersaVid-R1Base Model=Qwen2.5-VL-7B2025.11 | 62.9 | — |