Long Video Understanding and Reasoning on VideoMME w/ subtitle
81.3AccuracyGemini 1.5 Pro
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Gemini 1.5 ProReasoning Prompt=false, Tool Calling=false2025.11 | 81.3 | 55.2 | |
| GPT-4oReasoning Prompt=false, Tool Calling=false2025.11 | 77.2 | 51.5 | |
| LongVT-7B-RFTReasoning Prompt=true, Tool Calling=true, Frame Sampling Regime=Dense2025.11 | 67 | 47.7 | |
| LongVT-7B-RLReasoning Prompt=true, Tool Calling=true, Frame Sampling Regime=Sparse2025.11 | 66.1 | 43.7 | |
| LongVT-7B-RLReasoning Prompt=true, Tool Calling=true, Frame Sampling Regime=Dense2025.11 | 66.1 | 46.6 | |
| LongVT-7B-SFTReasoning Prompt=true, Tool Calling=true, Frame Sampling Regime=Dense2025.11 | 64.9 | 44.1 | |
| Qwen2.5-VL-7BReasoning Prompt=false, Tool Calling=false, Frame Sampling Regime=Dense2025.11 | 64.3 | 46 | |
| Qwen2.5-VL-7BReasoning Prompt=false, Tool Calling=false, Frame Sampling Regime=Sparse2025.11 | 62.6 | 37.2 | |
| Video-R1-7BReasoning Prompt=true, Tool Calling=false, Frame Sampling Regime=Sparse2025.11 | 61 | 42.6 | |
| Video-Thinker-7BReasoning Prompt=true, Tool Calling=false, Frame Sampling Regime=Sparse2025.11 | 61 | 42.6 | |
| VideoRFT-7BReasoning Prompt=true, Tool Calling=false, Frame Sampling Regime=Sparse2025.11 | 60.9 | 42.3 | |
| Video-Thinker-7BReasoning Prompt=true, Tool Calling=false, Frame Sampling Regime=Dense2025.11 | 60.8 | 42.9 | |
| Video-R1-7BReasoning Prompt=true, Tool Calling=false, Frame Sampling Regime=Dense2025.11 | 60.5 | 42.7 | |
| VideoRFT-7BReasoning Prompt=true, Tool Calling=false, Frame Sampling Regime=Dense2025.11 | 49.2 | 37 | |
| LongVT-7B-SFTReasoning Prompt=true, Tool Calling=true, Frame Sampling Regime=Sparse2025.11 | 12.5 | 36.2 |