Offline Video Understanding on EgoSchema v1 (test)
72.2AccuracyGPT4-o
Evaluation Results
| Method | Links | |
|---|---|---|
| GPT4-oSize=–, # Frames (Max)=1fps, Budget |M|=–2026.05 | 72.2 | |
| CoRDSSize=7B, # Frames (Max)=768, Budget |M|=6K, Backbone=Qwen2-VL2026.05 | 68.4 | |
| CoRDSSize=7B, # Frames (Max)=768, Budget |M|=3K, Backbone=Qwen2-VL2026.05 | 68.1 | |
| LongVUSize=7B, # Frames (Max)=1fps, Budget |M|=8K2026.05 | 67.6 | |
| LLaVA-Next-Video (Original)Size=7B, # Frames (Max)=128, Budget |M|=25K, Backbone=LLaVA-Next-Video2026.05 | 67.6 | |
| Qwen2-VL + StreamMemSize=7B, # Frames (Max)=4.0/0.5 fps, Budget |M|=6K, Backbone=Qwen2-VL2026.05 | 67.2 | |
| CoRDSSize=7B, # Frames (Max)=768, Budget |M|=6K, Backbone=Qwen2.5-VL2026.05 | 66.4 | |
| CoRDSSize=7B, # Frames (Max)=128, Budget |M|=6K, Backbone=LLaVA-Next-Video2026.05 | 66.2 | |
| CoRDSSize=7B, # Frames (Max)=128, Budget |M|=3K, Backbone=LLaVA-Next-Video2026.05 | 66 | |
| LLaVA-Next-Video + InfiniPot-VSize=7B, # Frames (Max)=128, Budget |M|=6K, Backbone=LLaVA-Next-Video2026.05 | 65.8 | |
| Qwen2-VL + InfiniPot-VSize=7B, # Frames (Max)=768, Budget |M|=6K, Backbone=Qwen2-VL2026.05 | 65.6 | |
| Qwen2-VL (Original)Size=7B, # Frames (Max)=768, Budget |M|=50K, Backbone=Qwen2-VL2026.05 | 65.2 | |
| CoRDSSize=7B, # Frames (Max)=768, Budget |M|=3K, Backbone=Qwen2.5-VL2026.05 | 64.8 | |
| Qwen2.5-VL (Original)Size=3B, # Frames (Max)=768, Budget |M|=50K, Backbone=Qwen2.5-VL2026.05 | 64.4 | |
| CoRDSSize=3B, # Frames (Max)=768, Budget |M|=6K, Backbone=Qwen2.5-VL2026.05 | 64.1 | |
| CoRDSSize=3B, # Frames (Max)=768, Budget |M|=3K, Backbone=Qwen2.5-VL2026.05 | 63.4 | |
| Qwen2.5-VL (Original)Size=7B, # Frames (Max)=768, Budget |M|=50K, Backbone=Qwen2.5-VL2026.05 | 62.8 | |
| Qwen2.5-VL + StreamMemSize=3B, # Frames (Max)=4.0/0.5 fps, Budget |M|=6K, Backbone=Qwen2.5-VL2026.05 | 62.2 | |
| Qwen2.5-VL + InfiniPot-VSize=3B, # Frames (Max)=768, Budget |M|=6K, Backbone=Qwen2.5-VL2026.05 | 61.8 | |
| LLaVA-OVSize=7B, # Frames (Max)=32, Budget |M|=8K2026.05 | 60.1 | |
| LongVUSize=3B, # Frames (Max)=1fps, Budget |M|=8K2026.05 | 59.1 | |
| GPT4-VSize=–, # Frames (Max)=1fps, Budget |M|=–2026.05 | 55.6 |