Video Understanding on VideoMME, LongVideoBench, and MLVU
59.97Average ScoreLLaVA-OV-7B
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| LLaVA-OV-7BFLOPs (T)=48.8, FLOPs Ratio=100.0%, Retention Ratio=100%, Backbone=LLaVA-OneVision-7B2026.06 | 59.97 | 100 | |
| InfoMergeFLOPs (T)=6.3, FLOPs Ratio=12.9%, Retention Ratio=15%, Backbone=LLaVA-OneVision-7B2026.06 | 59.23 | 98.8 | |
| FastVIDFLOPs (T)=6.3, FLOPs Ratio=12.9%, Retention Ratio=15%, Backbone=LLaVA-OneVision-7B2026.06 | 58.97 | 98.3 | |
| MMG-VidFLOPs (T)=6.3, FLOPs Ratio=12.9%, Retention Ratio=15%, Backbone=LLaVA-OneVision-7B2026.06 | 58.47 | 97.5 | |
| FastVIDFLOPs (T)=4.2, FLOPs Ratio=8.5%, Retention Ratio=10%, Backbone=LLaVA-OneVision-7B2026.06 | 58.2 | 97.1 | |
| InfoMergeFLOPs (T)=4.2, FLOPs Ratio=8.5%, Retention Ratio=10%, Backbone=LLaVA-OneVision-7B2026.06 | 58.2 | 97.1 | |
| VisionZipFLOPs (T)=6.3, FLOPs Ratio=12.9%, Retention Ratio=15%, Backbone=LLaVA-OneVision-7B2026.06 | 57.83 | 96.4 | |
| VisionZipFLOPs (T)=4.2, FLOPs Ratio=8.5%, Retention Ratio=10%, Backbone=LLaVA-OneVision-7B2026.06 | 56.7 | 94.5 | |
| FastVFLOPs (T)=9.4, FLOPs Ratio=19.2%, Retention Ratio=100%/15%, Backbone=LLaVA-OneVision-7B2026.06 | 56.43 | 94.1 | |
| InfoMergeFLOPs (T)=2.1, FLOPs Ratio=4.2%, Retention Ratio=5%, Backbone=LLaVA-OneVision-7B2026.06 | 55.4 | 92.4 | |
| FastVFLOPs (T)=7.4, FLOPs Ratio=15.1%, Retention Ratio=100%/10%, Backbone=LLaVA-OneVision-7B2026.06 | 55 | 91.7 | |
| FastVIDFLOPs (T)=2.1, FLOPs Ratio=4.2%, Retention Ratio=5%, Backbone=LLaVA-OneVision-7B2026.06 | 54.6 | 91 | |
| Vanilla#Retained Tokens=62722025.12 | 54.1 | 100 | |
| FastV + DyToK#Retained Tokens=4704 (↓ 25%)2025.12 | 54 | 99.8 | |
| DyCoke + DyToK#Retained Tokens=4704 (↓ 25%)2025.12 | 54 | 99.8 | |
| FastV + DyToK#Retained Tokens=3136 (↓ 50%)2025.12 | 54 | 99.8 | |
| DyCoke#Retained Tokens=4704 (↓ 25%)2025.12 | 53.8 | 99.4 | |
| DyCoke + DyToK#Retained Tokens=3136 (↓ 50%)2025.12 | 53.8 | 99.4 | |
| FastV#Retained Tokens=4704 (↓ 25%)2025.12 | 53.7 | 99.3 | |
| DyCoke#Retained Tokens=3136 (↓ 50%)2025.12 | 53.5 | 98.9 | |
| FastV + DyToK#Retained Tokens=1568 (↓ 75%)2025.12 | 53.3 | 98.5 | |
| VisionZipFLOPs (T)=2.1, FLOPs Ratio=4.2%, Retention Ratio=5%, Backbone=LLaVA-OneVision-7B2026.06 | 53.1 | 88.5 | |
| FastV#Retained Tokens=3136 (↓ 50%)2025.12 | 53 | 98 | |
| DyCoke + DyToK#Retained Tokens=1568 (↓ 75%)2025.12 | 52.4 | 96.9 | |
| FastV#Retained Tokens=1568 (↓ 75%)2025.12 | 52.1 | 96.3 | |
| DyCoke#Retained Tokens=1568 (↓ 75%)2025.12 | 51.8 | 95.7 | |
| FastVFLOPs (T)=5.4, FLOPs Ratio=11.1%, Retention Ratio=100%/5%, Backbone=LLaVA-OneVision-7B2026.06 | 51.77 | 86.3 | |
| FastV + DyToK#Retained Tokens=896 (↓ 85%)2025.12 | 50.2 | 92.8 | |
| DyCoke + DyToK#Retained Tokens=896 (↓ 85%)2025.12 | 50.1 | 92.6 | |
| DyCoke#Retained Tokens=896 (↓ 85%)2025.12 | 49.8 | 92.1 | |
| DyCoke + DyToK#Retained Tokens=448 (↓ 90%)2025.12 | 48.6 | 89.8 | |
| DyCoke#Retained Tokens=448 (↓ 90%)2025.12 | 48.4 | 89.5 | |
| FastV#Retained Tokens=896 (↓ 85%)2025.12 | 46.9 | 86.7 |