Extremely long-video understanding on LVBench
78.7ScoreGemini-2.5-Pro
Evaluation results
| Method | Links | |
|---|---|---|
| Gemini-2.5-ProLLM Size=-2026.03 | 78.7 | |
| Gemini-1.5-ProLLM Size=-2026.03 | 65.7 | |
| Qwen3-VL w/ AdaptTokenLLM Size=8B2026.03 | 60.6 | |
| Qwen2.5-VL w/ AdaptTokenLLM Size=72B2026.03 | 59.7 | |
| Qwen3-VLLLM Size=8B2026.03 | 58 | |
| Qwen2.5-VL w/ FlexSelectLLM Size=72B2026.03 | 56.6 | |
| Qwen2.5-VL w/ AdaptTokenLLM Size=7B2026.03 | 54.8 | |
| Qwen2.5-VL w/ AdaptToken-LiteLLM Size=7B2026.03 | 53.3 | |
| Qwen2.5-VL w/ AdaReTAKELLM Size=72B2026.03 | 53.3 | |
| InternVL2.5 w/ AdaptTokenLLM Size=8B2026.03 | 52.1 | |
| InternVL2.5 w/ ZoomVLLM Size=8B2026.03 | 51.5 | |
| InternVL2.5 w/ AdaptToken-LiteLLM Size=8B2026.03 | 51.3 | |
| Qwen2.5-VL w/ ZoomVLLM Size=7B2026.03 | 51.3 | |
| Qwen2.5-VL w/ AdaReTAKELLM Size=7B2026.03 | 51.2 | |
| Qwen2.5-VL w/ FlexSelectLLM Size=7B2026.03 | 51.2 | |
| InternVL2.5 w/ FlexSelectLLM Size=8B2026.03 | 49.7 | |
| Qwen2.5-VLLLM Size=72B2026.03 | 47.3 | |
| InternVL2.5 w/ SeViCESLLM Size=8B2026.03 | 46.7 | |
| InternVL2.5 w/ TriumphLLM Size=8B2026.03 | 46.6 | |
| Qwen2.5-VL w/ SeViCESLLM Size=7B2026.03 | 45.4 | |
| VideoLLaMA3LLM Size=7B2026.03 | 45.3 | |
| Qwen2.5-VLLLM Size=7B2026.03 | 45.3 | |
| ViLAMPLLM Size=7B2026.03 | 45.2 | |
| InternVL2.5LLM Size=8B2026.03 | 43.4 | |
| OursSize=7B, #Frames=322026.06 | 40.5 | |
| Ours(wo-mv)Size=7B, #Frames=322026.06 | 40.4 | |
| KangarooSize=8B, #Frames=642026.06 | 39.4 | |
| Video-R1Size=7B, #Frames=322026.06 | 38.7 | |
| LLaVA-OneVisionSize=7B, #Frames=642026.06 | 38.4 | |
| Qwen2.5-VL-7B-SFTSize=7B, #Frames=322026.06 | 35.7 | |
| Video-R1-wo-imageSize=7B, #Frames=322026.06 | 35.6 | |
| Video-R1-wo-imageSize=7B, #Frames=162026.06 | 34.7 | |
| VideoChat-R1Size=7B, #Frames=162026.06 | 34.1 | |
| Qwen2.5-VL-7BSize=7B2026.06 | 32.6 | |
| GPT-4oLLM Size=-2026.03 | 27 |