Video Understanding on Video-MME
78.78Overall ScoreFull-precision
Evaluation results
| Method | Links | |
|---|---|---|
| Full-precisionModel=Qwen3-VL-8B, Quantization=Full-precision2025.10 | 78.78 | |
| Full-precisionModel=Qwen3-VL-2B, Quantization=Full-precision2025.10 | 73.44 | |
| EgoVITAModel Family=Qwen3-VL Family2025.11 | 72.2 | |
| EgoVITA♢Model Family=Self-Trained2025.11 | 71.5 | |
| Qwen3-VLModel Family=Qwen3-VL Family2025.11 | 71.4 | |
| SFTModel Family=Qwen3-VL Family2025.11 | 71.2 | |
| GRPO (Format + Answer)Model Family=Qwen3-VL Family2025.11 | 71.1 | |
| EgoVITAModel Family=InternVL-3.5 Family2025.11 | 68.3 | |
| InternVL-3.5Model Family=InternVL-3.5 Family2025.11 | 65.6 | |
| FLoCModel=InternVL3-8B, Compression Ratio=2^-32025.10 | 64.93 | |
| LongVUModel=InternVL3-8B, Compression Ratio=2^-32025.10 | 64.7 | |
| EgoVITAModel Family=Qwen2.5-VL Family2025.11 | 64.6 | |
| DivPruneModel=InternVL3-8B, Compression Ratio=2^-32025.10 | 64.07 | |
| DyCokeModel=InternVL3-8B, Compression Ratio=2^-32025.10 | 63.96 | |
| STTMModel=InternVL3-8B, Compression Ratio=2^-32025.10 | 63.52 | |
| FLoCModel=InternVL3-8B, Compression Ratio=2^-42025.10 | 63.41 | |
| FLoCModel=Qwen2.5-VL-7B, Compression Ratio=2^-32025.10 | 63.33 | |
| GRPO (Format + Answer)Model Family=InternVL-3.5 Family2025.11 | 63.2 | |
| HumanOmni2025.08 | 63.1 | |
| GRPO (Format + Answer)Model Family=Qwen2.5-VL Family2025.11 | 62.9 | |
| Qwen2.5-OmniRL Method=AVATAR2025.08 | 62.8 | |
| TS-LLaVAModel=InternVL3-8B, Compression Ratio=2^-32025.10 | 62.78 | |
| Qwen2.5-VLModel Family=Qwen2.5-VL Family2025.11 | 62.2 | |
| LongVUModel=Qwen2.5-VL-7B, Compression Ratio=2^-32025.10 | 62.19 | |
| DyCokeModel=Qwen2.5-VL-7B, Compression Ratio=2^-32025.10 | 62.11 | |
| SFTModel Family=InternVL-3.5 Family2025.11 | 62.1 | |
| DivPruneModel=InternVL3-8B, Compression Ratio=2^-42025.10 | 61.93 | |
| DivPruneModel=Qwen2.5-VL-7B, Compression Ratio=2^-32025.10 | 61.63 | |
| Ola-7BRL Method=AVATAR2025.08 | 61.4 | |
| DyCokeModel=InternVL3-8B, Compression Ratio=2^-42025.10 | 61.37 | |
| TS-LLaVAModel=Qwen2.5-VL-7B, Compression Ratio=2^-32025.10 | 61.15 | |
| ScissorModel=InternVL3-8B, Compression Ratio=2^-32025.10 | 61.15 | |
| FastVIDModel=Qwen2.5-VL-7B, Compression Ratio=2^-32025.10 | 60.89 | |
| FLoCModel=Qwen2.5-VL-7B, Compression Ratio=2^-42025.10 | 60.89 | |
| DivPruneModel=InternVL3-8B, Compression Ratio=2^-52025.10 | 60.85 | |
| FLoCModel=InternVL3-8B, Compression Ratio=2^-52025.10 | 60.81 | |
| Echolnk2025.08 | 60.8 | |
| Omni-R12025.08 | 60.7 | |
| RandomModel=InternVL3-8B, Compression Ratio=2^-32025.10 | 60.59 | |
| Qwen2.5-OmniRL Method=GRPO2025.08 | 60.5 | |
| RandomModel=Qwen2.5-VL-7B, Compression Ratio=2^-32025.10 | 60.3 | |
| Ola-7BRL Method=GRPO2025.08 | 60.2 | |
| STTMModel=InternVL3-8B, Compression Ratio=2^-42025.10 | 60.15 | |
| SFTModel Family=Qwen2.5-VL Family2025.11 | 60.1 | |
| Qwen2.5-VL2025.08 | 60.1 | |
| TW-GRPO2025.08 | 60.1 | |
| VideoRFT2025.08 | 59.8 | |
| RandomModel=InternVL3-8B, Compression Ratio=2^-42025.10 | 59.74 | |
| TS-LLaVAModel=InternVL3-8B, Compression Ratio=2^-42025.10 | 59.63 | |
| STTMModel=Qwen2.5-VL-7B, Compression Ratio=2^-32025.10 | 59.52 | |
| Video-R12025.08 | 59.3 | |
| DyCokeModel=InternVL3-8B, Compression Ratio=2^-52025.10 | 59.22 | |
| Ola-7BRL Method=Baseline2025.08 | 59.1 | |
| ScissorModel=InternVL3-8B, Compression Ratio=2^-42025.10 | 58.89 | |
| TS-LLaVAModel=InternVL3-8B, Compression Ratio=2^-52025.10 | 58.89 | |
| DivPruneModel=Qwen2.5-VL-7B, Compression Ratio=2^-42025.10 | 58.85 | |
| TS-LLaVAModel=Qwen2.5-VL-7B, Compression Ratio=2^-42025.10 | 58.78 | |
| FastVIDModel=Qwen2.5-VL-7B, Compression Ratio=2^-42025.10 | 58.67 | |
| FLoCModel=Qwen2.5-VL-7B, Compression Ratio=2^-52025.10 | 58.63 | |
| ScissorModel=Qwen2.5-VL-7B, Compression Ratio=2^-32025.10 | 58.59 | |
| Qwen2.5-OmniRL Method=Baseline2025.08 | 58.3 | |
| PruneVidModel=Qwen2.5-VL-7B, Compression Ratio=2^-32025.10 | 58.19 | |
| LongVUModel=Qwen2.5-VL-7B, Compression Ratio=2^-42025.10 | 58.07 | |
| STTMModel=InternVL3-8B, Compression Ratio=2^-52025.10 | 57.52 | |
| RandomModel=Qwen2.5-VL-7B, Compression Ratio=2^-42025.10 | 57.44 | |
| PruneVidModel=InternVL3-8B, Compression Ratio=2^-32025.10 | 57.41 | |
| RandomModel=InternVL3-8B, Compression Ratio=2^-52025.10 | 57.3 | |
| FastVIDModel=Qwen2.5-VL-7B, Compression Ratio=2^-52025.10 | 57.19 | |
| STTMModel=Qwen2.5-VL-7B, Compression Ratio=2^-42025.10 | 57.15 | |
| DyCokeModel=Qwen2.5-VL-7B, Compression Ratio=2^-42025.10 | 57 | |
| AV-Reasoner2025.08 | 56.8 | |
| LongVUModel=InternVL3-8B, Compression Ratio=2^-42025.10 | 56.48 | |
| ScissorModel=InternVL3-8B, Compression Ratio=2^-52025.10 | 56.33 | |
| LongVUModel=InternVL3-8B, Compression Ratio=2^-52025.10 | 55.96 | |
| DivPruneModel=Qwen2.5-VL-7B, Compression Ratio=2^-52025.10 | 55.78 | |
| RandomModel=Qwen2.5-VL-7B, Compression Ratio=2^-52025.10 | 55.56 | |
| ScissorModel=Qwen2.5-VL-7B, Compression Ratio=2^-42025.10 | 55.26 | |
| STTMModel=Qwen2.5-VL-7B, Compression Ratio=2^-52025.10 | 55.26 | |
| TS-LLaVAModel=Qwen2.5-VL-7B, Compression Ratio=2^-52025.10 | 55.07 | |
| DyCokeModel=Qwen2.5-VL-7B, Compression Ratio=2^-52025.10 | 54.37 | |
| PruneVidModel=Qwen2.5-VL-7B, Compression Ratio=2^-42025.10 | 54.11 | |
| PruneVidModel=InternVL3-8B, Compression Ratio=2^-42025.10 | 53.81 | |
| LongVUModel=Qwen2.5-VL-7B, Compression Ratio=2^-52025.10 | 53.41 | |
| RegCacheModel=Qwen3-VL-8B, Quantization=4-bit, Base PTQ=RepQ-ViT2025.10 | 52.89 | |
| ScissorModel=Qwen2.5-VL-7B, Compression Ratio=2^-52025.10 | 51.89 | |
| PruneVidModel=InternVL3-8B, Compression Ratio=2^-52025.10 | 51.41 | |
| PruneVidModel=Qwen2.5-VL-7B, Compression Ratio=2^-52025.10 | 51.11 | |
| RepQ-ViTModel=Qwen3-VL-8B, Quantization=4-bit2025.10 | 50.44 | |
| RegCacheModel=Qwen3-VL-2B, Quantization=4-bit, Base PTQ=RepQ-ViT2025.10 | 49.22 | |
| RepQ-ViTModel=Qwen3-VL-2B, Quantization=4-bit2025.10 | 44.56 | |
| NaiveModel=Qwen3-VL-8B, Quantization=4-bit2025.10 | 35.33 | |
| NaiveModel=Qwen3-VL-2B, Quantization=4-bit2025.10 | 32.11 |