Video Understanding on LongVideoBench (Accuracy Delta)
66.7AccuracyGPT-4o
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| GPT-4oModel Category=Proprietary Models, Venue=-2026.03 | 66.7 | — | — | — | |
| Gemini 1.5 proModel Category=Proprietary Models, Venue=-2026.03 | 64 | — | — | — | |
| AnchorAtten.Backbone=Qwen2.5-VL-7B-Instruct, #FPS=1, #Tokens=26K, Average Sparsity=74.62026.03 | 60.8 | — | — | — | |
| ERABackbone=LLaVA-Video-7B, Token Retention Ratio=25%, Pruning Transfer Protocol=image-only pruning transferred to video inputs2026.06 | 60.2 | — | — | — | |
| XAtten.Backbone=InternVL-3.5-8B, #Frames=64, #Tokens=17K, Average Sparsity=78.12026.03 | 59.9 | — | — | — | |
| Full Atten.Backbone=Qwen2.5-VL-7B-Instruct, #FPS=1, #Tokens=26K, Average Sparsity=0.02026.03 | 59.9 | — | — | — | |
| Penguin-VLParameters=2B2026.03 | 59.5 | — | — | — | |
| Full Atten.Backbone=InternVL-3.5-8B, #Frames=64, #Tokens=17K, Average Sparsity=0.02026.03 | 59.4 | — | — | — | |
| AnchorAtten.Backbone=InternVL-3.5-8B, #Frames=64, #Tokens=17K, Average Sparsity=78.62026.03 | 59.4 | — | — | — | |
| VecAtten.Backbone=Qwen2.5-VL-7B-Instruct, #FPS=1, #Tokens=26K, Average Sparsity=78.52026.03 | 59.4 | — | — | — | |
| FlexPref.Backbone=InternVL-3.5-8B, #Frames=64, #Tokens=17K, Average Sparsity=76.52026.03 | 59 | — | — | — | |
| VecAtten.Backbone=InternVL-3.5-8B, #Frames=64, #Tokens=17K, Average Sparsity=78.62026.03 | 59 | — | — | — | |
| InternVL3-8BFrames=322026.05 | 58.7 | — | — | — | |
| XAtten.Backbone=Qwen2.5-VL-7B-Instruct, #FPS=1, #Tokens=26K, Average Sparsity=73.62026.03 | 58.5 | — | — | — | |
| InternVL3-8B + LiteFrameFrames=256 (8×)2026.05 | 58.3 | — | — | — | |
| GPT-4oFrame=322026.03 | 58.2 | — | — | — | |
| LongVILA-R1-7BModel Category=Open-source Offline Models, Venue=NeurIPS’252026.03 | 58 | — | — | — | |
| VST-7BModel Category=Open-source Online Models, Venue=-2026.03 | 58 | — | — | — | |
| REVISOR-7BModel Category=Open-source Offline Models, Venue=CVPR’262026.03 | 57.5 | — | — | — | |
| InternVL3.5Parameters=2B2026.03 | 57.4 | — | — | — | |
| InternVL3-8B + LiteFrameFrames=128 (8×)2026.05 | 57.2 | — | — | — | |
| FlexPref.Backbone=Qwen2.5-VL-7B-Instruct, #FPS=1, #Tokens=26K, Average Sparsity=73.62026.03 | 56.7 | — | — | — | |
| DyCokeRetention Ratio R=32.5%2026.05 | 56.6 | — | — | — | |
| FastVRetention Ratio R=100%/20%2026.05 | 56.6 | — | — | — | |
| LLaVA-OneVision-7BBackbone=LLaVA-OneVision-7B, Token Retention Ratio=100%2026.06 | 56.6 | — | — | — | |
| InternVL3-8BFrames=162026.05 | 56.5 | — | — | — | |
| LLaVA-OV 7BRetention Ratio R=100%2026.05 | 56.4 | — | — | — | |
| LongVA-7BModel Category=Open-source Offline Models, Venue=TMLR’252026.03 | 56.3 | — | — | — | |
| DynaTokRetention Ratio R=25%2026.05 | 56.3 | — | — | — | |
| DynaTokRetention Ratio R=19.9%2026.05 | 55.7 | — | — | — | |
| InternVL3-8B + FastVIDFrames=128 (4×)2026.05 | 55.6 | — | — | — | |
| ERABackbone=LLaVA-OneVision-7B, Token Retention Ratio=25%, Pruning Transfer Protocol=image-only pruning transferred to video inputs2026.06 | 55.6 | — | — | — | |
| FastVRetention Ratio R=100%/25%2026.05 | 55.5 | — | — | — | |
| TimeChatOnline-7BModel Category=Open-source Online Models, Venue=MM’252026.03 | 55.4 | — | — | — | |
| DynaTokRetention Ratio R=14.8%2026.05 | 55.4 | — | — | — | |
| DynaTokRetention Ratio R=9.7%2026.05 | 55.4 | — | — | — | |
| Gemini-1.5-ProFrame=322026.03 | 55.2 | — | — | — | |
| EVA-GRPOFrame=25.3*2026.03 | 55 | — | — | — | |
| FastVRetention Ratio R=100%/15%2026.05 | 54.9 | — | — | — | |
| InternVL3-8BFrames=82026.05 | 54.8 | — | — | — | |
| InternVL3-8B + LiteFrameFrames=64 (8×)2026.05 | 53.6 | — | — | — | |
| EVA-KTOFrame=35.6*2026.03 | 53.2 | — | — | — | |
| FrameThinkerFrame=21.12026.03 | 52.9 | — | — | — | |
| Video-R1Frame=322026.03 | 52.7 | — | — | — | |
| InternVL3-8B + FastVIDFrames=32 (4×)2026.05 | 52.6 | — | — | — | |
| InternVL3-8B + FastVIDFrames=64 (4×)2026.05 | 52.6 | — | — | — | |
| Qwen3-VLParameters=2B2026.03 | 52.1 | — | — | — | |
| FastVRetention Ratio R=100%/10%2026.05 | 52.1 | — | — | — | |
| VisionZipRetention Ratio R=25%2026.05 | 51.2 | — | — | — | |
| VisionZipRetention Ratio R=19.9%2026.05 | 50 | — | — | — | |
| EVA-SFTFrame=33.8*2026.03 | 49.9 | — | — | — | |
| SmolVLM2Parameters=2.2B2026.03 | 49.7 | — | — | — | |
| VideoChat-R1Frame=322026.03 | 49.1 | — | — | — | |
| DyCokeRetention Ratio R=25%2026.05 | 48.1 | — | — | — | |
| VisionZipRetention Ratio R=14.8%2026.05 | 46.9 | — | — | — | |
| VisionZipRetention Ratio R=9.7%2026.05 | 43.5 | — | — | — | |
| Qwen2.5-VLFrame=322026.03 | 43.2 | — | — | — | |
| Gemma3n E2B-itParameters=E2B-it, Inference template=Penguin's template, Frame budget=322026.03 | 43 | — | — | — | |
| ShareGPT4VideoFrame=162026.03 | 39.7 | — | — | — | |
| A.I.R.Model=Qwen2.5-VL [2], Size=7B, Frame=≤32§2026.02 | — | 58.1 | 61.4 | 3.3 | |
| A.I.R.Model=InternVL-3 [47], Size=8B, Frame=≤32§2026.02 | — | 58.3 | 62.8 | 4.5 | |
| AKSModel=LLaVA-OV [10], Size=7B, Frame=82026.02 | — | 54.2 | 58.4 | 4.2 | |
| AKSModel=LLaVA-Video [44], Size=7B, Frame=82026.02 | — | 54.9 | 59.6 | 4.7 | |
| AKSModel=Qwen2.5-VL [2], Size=7B, Frame=322026.02 | — | 58.9 | 63.2 | 4.3 | |
| AKSModel=InternVL-3 [47], Size=8B, Frame=322026.02 | — | 58.5 | 61.5 | 3 | |
| BOLTModel=LLaVA-OV [10], Size=7B, Frame=82026.02 | — | 54.2 | 55.6 | 1.4 | |
| FrameOracleModel=LLaVA-OV [10], Size=7B, Frame=82026.02 | — | 54.3 | 56 | 1.7 | |
| FrameOracleModel=LLaVA-Video [44], Size=7B, Frame=82026.02 | — | 54.2 | 56.9 | 2.7 | |
| KFCModel=LLaVA-OV [10], Size=7B, Frame=82026.02 | — | 54.5 | 55.6 | 1.1 | |
| KFCModel=LLaVA-Video [44], Size=7B, Frame=82026.02 | — | 54.2 | 56.5 | 2.3 | |
| MDP3Model=Qwen2.5-VL [2], Size=7B, Frame=322026.02 | — | 58.1 | 60 | 1.9 | |
| MDP3Model=InternVL-3 [47], Size=8B, Frame=322026.02 | — | 58.3 | 60.9 | 2.6 | |
| WFS-SBModel=LLaVA-OV [10], Size=7B, Frame=82026.02 | — | 54.2 | 59.8 | 5.6 | |
| WFS-SBModel=LLaVA-Video [44], Size=7B, Frame=82026.02 | — | 54.9 | 61.1 | 6.2 | |
| WFS-SBModel=Qwen2.5-VL [2], Size=7B, Frame=322026.02 | — | 58.9 | 64.4 | 5.5 | |
| WFS-SBModel=InternVL-3 [47], Size=8B, Frame=322026.02 | — | 58.5 | 62.9 | 4.4 |