Video Question Answering on MVBench (test)
73.6AccuracyQwen2-VL-72B
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Qwen2-VL-72BModel Group=Open-source (>10B), Parameters=72B2025.01 | 73.6 | — | |
| Previous SOTACitation=[20]2025.01 | 72 | — | |
| Tarsier2-7BParameters=7B2025.01 | 71.5 | — | |
| Tarsier-34BModel Group=Open-source (>10B), Parameters=34B2025.01 | 67.6 | — | |
| Qwen2-VL-7BModel Group=Open-source (<=10B), Parameters=7B2025.01 | 67 | — | |
| LLaVA-Video-72BModel Group=Open-source (>10B), Parameters=72B2025.01 | 64.1 | — | |
| Tarsier-7BModel Group=Open-source (<=10B), Parameters=7B2025.01 | 62.6 | — | |
| Qwen2.5-VL-7Baugmented_with=MASS2025.11 | 61.59 | — | |
| VideoChat2LLM Size=7B2024.10 | 60.4 | — | |
| VideoChat-TLLM Size=7B2024.10 | 59.9 | — | |
| LLaVA-OV-72BModel Group=Open-source (>10B), Parameters=72B2025.01 | 59.4 | — | |
| LLaVA-Video-7BModel Group=Open-source (<=10B), Parameters=7B2025.01 | 58.6 | — | |
| GPT-4o2025.11 | 58.5 | — | |
| GPT-4oModel Group=Proprietary2025.01 | 57.5 | — | |
| LLaVA-OV-7BModel Group=Open-source (<=10B), Parameters=7B2025.01 | 56.7 | — | |
| LLaVA-OneVision-7B2025.11 | 55.24 | — | |
| MovieChatLLM Size=7B2024.10 | 55.1 | — | |
| Gemini-2.5-Flash2025.11 | 54.32 | — | |
| Qwen2.5-VL-7B2025.11 | 51.3 | — | |
| LLaVA-OneVision-7Baugmented_with=MASS2025.11 | 48.33 | — | |
| InternVL3.5-8B-Flash2025.11 | 45.78 | — | |
| VanillaBase Model=LLaVA-NeXT-Video-7B, Retained Tokens=Full, Pruning Rate=0%2026.01 | 44.95 | 100 | |
| ViTCoPBase Model=LLaVA-NeXT-Video-7B, Retained Tokens=128, Pruning Rate=88.9%2026.01 | 43.3 | 96.3 | |
| SparseVLMBase Model=LLaVA-NeXT-Video-7B, Retained Tokens=128, Pruning Rate=88.9%2026.01 | 43.13 | 96 | |
| LLAMA-VidLLM Size=7B2024.10 | 41.9 | — | |
| VisionZipBase Model=LLaVA-NeXT-Video-7B, Retained Tokens=128, Pruning Rate=88.9%2026.01 | 40.8 | 90.8 | |
| FastVBase Model=LLaVA-NeXT-Video-7B, Retained Tokens=128, Pruning Rate=88.9%2026.01 | 40.78 | 90.7 | |
| PyramidDropBase Model=LLaVA-NeXT-Video-7B, Retained Tokens=128, Pruning Rate=88.9%2026.01 | 38.8 | 86.3 | |
| TimeChatLLM Size=7B2024.10 | 38.5 | — | |
| LLaVA-NeXT#Frames=4, Selection Strategy=GCMI (Query-based)2026.01 | 37.22 | — | |
| Video-LLaVA#Frames=4, Selection Strategy=GCMI (Query-based)2026.01 | 37.1 | — | |
| LLaVA-NeXT#Frames=12, Selection Strategy=GCMI (Query-based)2026.01 | 36.7 | — | |
| Video-LLaVA#Frames=12, Selection Strategy=GCMI (Query-based)2026.01 | 36.45 | — | |
| Video-LLaVA#Frames=12, Selection Strategy=FLMI (Query-based)2026.01 | 36.08 | — | |
| Video-LLaVA#Frames=4, Selection Strategy=FLMI (Query-based)2026.01 | 35.13 | — | |
| LLaVA-NeXT#Frames=4, Selection Strategy=AKS2026.01 | 34.84 | — | |
| LLaVA-NeXT#Frames=4, Selection Strategy=FLMI (Query-based)2026.01 | 34.32 | — | |
| LLaVA-NeXT#Frames=12, Selection Strategy=FLMI (Query-based)2026.01 | 34.11 | — | |
| Video-LLaVA#Frames=12, Selection Strategy=AKS2026.01 | 33.4 | — | |
| LLaVA-NeXT#Frames=12, Selection Strategy=AKS2026.01 | 33.28 | — | |
| Video-LLaVA#Frames=4, Selection Strategy=Uniform2026.01 | 33.21 | — | |
| LLaVA-NeXT#Frames=12, Selection Strategy=Uniform2026.01 | 32.82 | — | |
| Video-LLaVA#Frames=4, Selection Strategy=AKS2026.01 | 32.62 | — | |
| Video-LLaVA#Frames=12, Selection Strategy=Uniform2026.01 | 32.38 | — | |
| LLaVA-NeXT#Frames=4, Selection Strategy=Uniform2026.01 | 31.48 | — |