Long Video Understanding on LongVideo Bench (Summary Metrics)
62.8ScoreQwen3-VL-8B-Instruct
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Qwen3-VL-8B-InstructMax Input Frames=64, Retention Ratio=100.0%2026.03 | 62.8 | — | |
| FlashVIDBackbone=Qwen2.5VL-7B, Frames / Pruning Configuration=F160 (5×) / 20%2026.05 | 61.6 | — | |
| V-CASTMax Input Frames=64, Retention Ratio=25%2026.03 | 61.2 | — | |
| EchoPruneBackbone=Qwen2.5VL-7B, Frames / Pruning Configuration=F160 (5×) / 20%2026.05 | 61.1 | — | |
| RandomBackbone=Qwen2.5VL-7B, Frames / Pruning Configuration=F160 (5×) / 20%2026.05 | 60.5 | — | |
| FastVIDMax Input Frames=64, Retention Ratio=25%2026.03 | 60.4 | — | |
| Qwen3-VL-8B-InstructMax Input Frames=32, Retention Ratio=100.0%2026.03 | 60.3 | — | |
| FastVIDMax Input Frames=64, Retention Ratio=15%2026.03 | 59.7 | — | |
| VidCom2Max Input Frames=64, Retention Ratio=25%2026.03 | 59.6 | — | |
| V-CASTMax Input Frames=64, Retention Ratio=15%2026.03 | 59.6 | — | |
| MI-PrunerBackbone=Qwen2.5VL-7B, Frames / Pruning Configuration=F160 (5×) / 20%2026.05 | 59.3 | — | |
| VisionZipMax Input Frames=64, Retention Ratio=25%2026.03 | 58.9 | — | |
| HoliTomMax Input Frames=64, Retention Ratio=25%2026.03 | 58.9 | — | |
| Qwen2.5VL-7BBackbone=Qwen2.5VL-7B, Frames / Pruning Configuration=F32 (1×) / 100%2026.05 | 58.9 | — | |
| FlashVIDMax Input Frames=32, Retention Ratio=25%2026.03 | 58.8 | — | |
| FastVIDMax Input Frames=32, Retention Ratio=25%2026.03 | 58.7 | — | |
| LLaVA-Video-7BRetention Ratio γ=100%2026.04 | 58.7 | — | |
| HoliTomMax Input Frames=64, Retention Ratio=15%2026.03 | 58.3 | — | |
| V-CASTMax Input Frames=32, Retention Ratio=25%2026.03 | 58.2 | — | |
| KiTokeRetention Ratio γ=25%2026.04 | 58.2 | — | |
| VidCom2Max Input Frames=32, Retention Ratio=25%2026.03 | 58 | — | |
| FlashVIDMax Input Frames=32, Retention Ratio=15%2026.03 | 57.8 | — | |
| VisionZipMax Input Frames=64, Retention Ratio=15%2026.03 | 57.7 | — | |
| V-CASTMax Input Frames=32, Retention Ratio=15%2026.03 | 57.7 | — | |
| PruneVIDRetention Ratio γ=25%2026.04 | 57.7 | — | |
| OursRetention Ratio=15%, Backbone=LLaVA-OneVision-7B2026.03 | 57.5 | 57.9 | |
| OursRetention Ratio=15%/7.5%2026.03 | 57.5 | — | |
| Ours (w/o M)Retention Ratio=10%2026.03 | 57.4 | — | |
| VidCom2Max Input Frames=64, Retention Ratio=15%2026.03 | 57.4 | — | |
| FastVRetention Ratio γ=100%/25%2026.04 | 57.4 | — | |
| FastVIDRetention Ratio γ=25%2026.04 | 57.3 | — | |
| FastVIDMax Input Frames=32, Retention Ratio=15%2026.03 | 57.1 | — | |
| KiTokeRetention Ratio γ=10%2026.04 | 57 | — | |
| HoliTomMax Input Frames=32, Retention Ratio=25%2026.03 | 56.8 | — | |
| VisionZipMax Input Frames=32, Retention Ratio=25%2026.03 | 56.7 | — | |
| VisionZipRetention Ratio=25%, Backbone=LLaVA-OneVision-7B2026.03 | 56.5 | 57.2 | |
| VidCom2Retention Ratio γ=25%2026.04 | 56.5 | — | |
| VanillaRetention Ratio=100%, Backbone=LLaVA-OneVision-7B2026.03 | 56.4 | 57.4 | |
| HoliTomRetention Ratio=15%, Backbone=LLaVA-OneVision-7B2026.03 | 56.4 | 57.3 | |
| LLaVA-OV-7BRetention Ratio=100%2026.03 | 56.4 | — | |
| HoliTomRetention Ratio=15%/7.5%2026.03 | 56.4 | — | |
| OursRetention Ratio=10%/5%2026.03 | 56.4 | — | |
| VisionZipRetention Ratio γ=25%2026.04 | 56.4 | — | |
| MI-PrunerBackbone=Qwen2.5VL-3B, Frames / Pruning Configuration=F160 (5×) / 20%2026.05 | 56.4 | — | |
| FastVidRetention Ratio=25%, Backbone=LLaVA-OneVision-7B2026.03 | 56.3 | 56.4 | |
| FastVIDRetention Ratio=10%2026.03 | 56.3 | — | |
| HoliTomRetention Ratio=10%/5%2026.03 | 56.3 | — | |
| FastVIDRetention Ratio=15%2026.03 | 56.2 | — | |
| VisionZipMax Input Frames=32, Retention Ratio=15%2026.03 | 56.2 | — | |
| DyCokeRetention Ratio γ=25%2026.04 | 56.2 | — | |
| VidCom2Max Input Frames=32, Retention Ratio=15%2026.03 | 56 | — | |
| FlashVIDBackbone=Qwen2.5VL-3B, Frames / Pruning Configuration=F160 (5×) / 20%2026.05 | 55.9 | — | |
| EchoPruneBackbone=Qwen2.5VL-3B, Frames / Pruning Configuration=F160 (5×) / 20%2026.05 | 55.9 | — | |
| HoliTomMax Input Frames=32, Retention Ratio=15%2026.03 | 55.8 | — | |
| FastVIDRetention Ratio γ=10%2026.04 | 55.7 | — | |
| KiTokeRetention Ratio γ=5%2026.04 | 55.7 | — | |
| PruneVIDRetention Ratio γ=10%2026.04 | 55.2 | — | |
| RandomBackbone=Qwen2.5VL-3B, Frames / Pruning Configuration=F160 (5×) / 20%2026.05 | 55.1 | — | |
| Qwen2.5VL-3BBackbone=Qwen2.5VL-3B, Frames / Pruning Configuration=F32 (1×) / 100%2026.05 | 54.6 | — | |
| OursRetention Ratio=5%/2.5%2026.03 | 54.5 | — | |
| VisionZipRetention Ratio=15%2026.03 | 54.4 | — | |
| DyToKRetention Ratio=15%2026.03 | 53.7 | — | |
| Ours (w/o M)Retention Ratio=5%2026.03 | 53.7 | — | |
| HoliTomRetention Ratio=5%/2.5%2026.03 | 53.6 | — | |
| FastVRetention Ratio γ=100%/10%2026.04 | 53.1 | — | |
| PruneVIDRetention Ratio γ=5%2026.04 | 52.5 | — | |
| DyToKRetention Ratio=10%2026.03 | 52.3 | — | |
| FsatVRetention Ratio=100%/10%2026.03 | 52.1 | — | |
| VisionZipRetention Ratio γ=10%2026.04 | 52.1 | — | |
| KiTokeRetention Ratio γ=1%2026.04 | 51.9 | — | |
| FastVIDRetention Ratio γ=5%2026.04 | 51.8 | — | |
| FastVIDRetention Ratio=5%2026.03 | 51.1 | — | |
| Ours (w/o M)Retention Ratio=2%2026.03 | 51 | — | |
| OursRetention Ratio=2%/1%2026.03 | 50.8 | — | |
| VidCom2Retention Ratio γ=10%2026.04 | 50.7 | — | |
| FastVRetention Ratio γ=100%/5%2026.04 | 50.3 | — | |
| VisionZipRetention Ratio=10%2026.03 | 49.3 | — | |
| DyToKRetention Ratio=5%2026.03 | 49.1 | — | |
| Ours (w/o M)Retention Ratio=1%2026.03 | 49.1 | — | |
| OursRetention Ratio=1%/0.5%2026.03 | 48.8 | — | |
| HoliTomRetention Ratio=2%/1%2026.03 | 48.5 | — | |
| FastVIDRetention Ratio γ=1%2026.04 | 48.2 | — | |
| HoliTomRetention Ratio=1%/0.5%2026.03 | 48.1 | — | |
| VisionZipRetention Ratio γ=5%2026.04 | 48.1 | — | |
| FastVRetention Ratio=100%/5%2026.03 | 47.9 | — | |
| PruneVIDRetention Ratio γ=1%2026.04 | 47.4 | — | |
| FastVIDRetention Ratio=2%2026.03 | 47.3 | — | |
| FastVRetention Ratio=100%/2%2026.03 | 47.1 | — | |
| VidCom2Retention Ratio γ=5%2026.04 | 47 | — | |
| VisionZipRetention Ratio=5%2026.03 | 46.4 | — | |
| FastVIDRetention Ratio=1%2026.03 | 46.1 | — | |
| FastVRetention Ratio γ=100%/1%2026.04 | 46.1 | — | |
| DyToKRetention Ratio=2%2026.03 | 45.9 | — | |
| VisionZipRetention Ratio γ=1%2026.04 | 45.9 | — | |
| DyToKRetention Ratio=1%2026.03 | 45.6 | — | |
| FastVRetention Ratio=100%/1%2026.03 | 45.5 | — | |
| VisionZipRetention Ratio=2%2026.03 | 45.1 | — | |
| VisionZipRetention Ratio=1%2026.03 | 44.4 | — | |
| VidCom2Retention Ratio γ=1%2026.04 | 44.4 | — | |
| FastVIDBase Model=Qwen3-VL-30B-A3B-Instruct, Retention Ratio=25%, Input Frames=642026.03 | — | 60.7 |