Long-video understanding on LVBench (test)
71.8LVBench ScoreSymphony
Evaluation Results
| Method | Links | |
|---|---|---|
| Symphony2026.03 | 71.8 | |
| DVDModel Category=Agent Based, reproduced=true2026.03 | 66.8 | |
| Qwen3-VL-235B-A22BFrames=7682026.03 | 63.6 | |
| MR. VideoModel Category=Others2026.03 | 60.8 | |
| Seed 1.6 VLModel Category=Open-Source VLMs, reproduced=true2026.03 | 58.1 | |
| OpenAI o3Model Category=Commercial VLMs2026.03 | 57.1 | |
| Qwen3-VL-4BModel=Qwen3-VL-4B, Pruning Strength=None2026.03 | 56.2 | |
| VideoDeepResearchModel Category=Agent Based2026.03 | 55.5 | |
| AdaRETAKEModel Category=Others2026.03 | 53.3 | |
| GPT-4oModel Category=Commercial VLMs2026.03 | 48.9 | |
| Qwen2.5-VL-72BModel Category=Open-Source VLMs2026.03 | 47.7 | |
| Qwen2.5-VL-72BFrames=7682026.03 | 47.3 | |
| PLM-8BModel=PLM-8B, Pruning Strength=None2026.03 | 44.5 | |
| Qwen3-VL-8B-Thinking + FrameRepeatFrames=128+82026.03 | 44.3 | |
| InternVL2.5-78BModel Category=Open-Source VLMs2026.03 | 43.6 | |
| InternVL3.5-8BModel=InternVL3.5-8B, Pruning Strength=None2026.03 | 43.4 | |
| STTSModel=STTS, Pruning Strength=30%2026.03 | 42.6 | |
| Qwen3-VL-8B-Thinking + CLIP-KFrames=128+82026.03 | 42.3 | |
| BaselineModel=Baseline, Pruning Strength=0%2026.03 | 42 | |
| Qwen3-VL-8B-Thinking + Random-KFrames=128+82026.03 | 41.6 | |
| Qwen3-VL-8B-Thinking + Bottom-KFrames=128+82026.03 | 41.2 | |
| ReAgent-VModel Category=Agent Based2026.03 | 41.2 | |
| STTSModel=STTS, Pruning Strength=40%2026.03 | 41.1 | |
| Qwen3-VL-8B-ThinkingFrames=1282026.03 | 40.6 | |
| STTSModel=STTS, Pruning Strength=50%2026.03 | 40.5 | |
| Qwen3-VL-8B-Thinking + FrameRepeatFrames=64+42026.03 | 39.3 | |
| Qwen3-VL-8B-Thinking + CLIP-KFrames=64+42026.03 | 38.4 | |
| Qwen2.5-VL-7B-Instruct + FrameRepeatFrames=128+82026.03 | 38.2 | |
| Qwen3-VL-8B-Thinking + Random-KFrames=64+42026.03 | 38.1 | |
| Qwen3-VL-8B-ThinkingFrames=642026.03 | 37.7 | |
| Qwen2.5-VL-7B-Instruct + CLIP-KFrames=128+82026.03 | 37.5 | |
| Qwen3-VL-8B-Thinking + Bottom-KFrames=64+42026.03 | 36.7 | |
| Qwen2.5-VL-7B-InstructFrames=1282026.03 | 36 | |
| Qwen2.5-VL-7B-Instruct + Random-KFrames=128+82026.03 | 36 | |
| Qwen2.5-VL-7B-Instruct + Bottom-KFrames=128+82026.03 | 35.4 | |
| Qwen2.5-VL-7B-Instruct + FrameRepeatFrames=64+42026.03 | 34.3 | |
| Qwen2.5-VL-7B-Instruct + CLIP-KFrames=64+42026.03 | 33.8 | |
| Qwen2.5-VL-7B-Instruct + Random-KFrames=64+42026.03 | 33.2 | |
| Gemini-1.5-ProModel Category=Commercial VLMs2026.03 | 33.1 | |
| Qwen2.5-VL-7B-InstructFrames=642026.03 | 32.9 | |
| Qwen2.5-VL-7B-Instruct + Bottom-KFrames=64+42026.03 | 32.3 | |
| VideoAgentModel Category=Agent Based2026.03 | 29.3 | |
| VideoTreeModel Category=Agent Based2026.03 | 28.8 |