Long Video Understanding and Reasoning on VideoMMMU
66Adaptation ScoreGPT-4o
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| GPT-4oReasoning Prompt=false, Tool Calling=false2025.11 | 66 | 62 | 55.7 | 51.5 | |
| Gemini 1.5 ProReasoning Prompt=false, Tool Calling=false2025.11 | 59 | 53.3 | 49.3 | 55.2 | |
| LongVT-7B-SFTReasoning Prompt=true, Tool Calling=true, Frame Sampling Regime=Sparse2025.11 | 37.7 | 46 | 58.3 | 36.2 | |
| VideoRFT-7BReasoning Prompt=true, Tool Calling=false, Frame Sampling Regime=Dense2025.11 | 37.7 | 40.7 | 48.7 | 37 | |
| Video-Thinker-7BReasoning Prompt=true, Tool Calling=false, Frame Sampling Regime=Dense2025.11 | 37.7 | 42.7 | 55.3 | 42.9 | |
| LongVT-7B-RLReasoning Prompt=true, Tool Calling=true, Frame Sampling Regime=Dense2025.11 | 37.7 | 42.3 | 56.3 | 46.6 | |
| Qwen2.5-VL-7BReasoning Prompt=false, Tool Calling=false, Frame Sampling Regime=Sparse2025.11 | 37.3 | 28 | 36.7 | 37.2 | |
| Video-R1-7BReasoning Prompt=true, Tool Calling=false, Frame Sampling Regime=Dense2025.11 | 37.3 | 38.7 | 46.3 | 42.7 | |
| VideoRFT-7BReasoning Prompt=true, Tool Calling=false, Frame Sampling Regime=Sparse2025.11 | 36.7 | 42 | 53 | 42.3 | |
| Video-R1-7BReasoning Prompt=true, Tool Calling=false, Frame Sampling Regime=Sparse2025.11 | 36.3 | 40.7 | 52.3 | 42.6 | |
| Qwen2.5-VL-7BReasoning Prompt=false, Tool Calling=false, Frame Sampling Regime=Dense2025.11 | 35.7 | 44.3 | 54.7 | 46 | |
| LongVT-7B-RFTReasoning Prompt=true, Tool Calling=true, Frame Sampling Regime=Dense2025.11 | 35.7 | 43.7 | 56.7 | 47.7 | |
| Video-Thinker-7BReasoning Prompt=true, Tool Calling=false, Frame Sampling Regime=Sparse2025.11 | 34.3 | 44.7 | 53 | 42.6 | |
| LongVT-7B-RLReasoning Prompt=true, Tool Calling=true, Frame Sampling Regime=Sparse2025.11 | 32.7 | 44.7 | 50 | 43.7 | |
| LongVT-7B-SFTReasoning Prompt=true, Tool Calling=true, Frame Sampling Regime=Dense2025.11 | 32.3 | 42 | 49.7 | 44.1 |