Long-video Understanding on OVOBench Real-Time
71.45AccuracyFull SA
Evaluation Results
| Method | Links | |
|---|---|---|
| Full SABackbone=InternVL3-8B2026.05 | 71.45 | |
| StateKVBackbone=InternVL3-8B, cache budget B=40962026.05 | 70.25 | |
| Full SABackbone=Eagle2.5-8B2026.05 | 69.3 | |
| StateKVBackbone=Eagle2.5-8B, cache budget B=40962026.05 | 68.7 | |
| Full SABackbone=QWen3-VL-8B2026.05 | 67.86 | |
| StateKVBackbone=QWen3-VL-8B, cache budget B=40962026.05 | 64.99 | |
| StateKVBackbone=QWen3-VL-4B, cache budget B=40962026.05 | 64.87 | |
| Full SABackbone=QWen3-VL-4B2026.05 | 64.76 | |
| StateKVBackbone=InternVL3-2B, cache budget B=40962026.05 | 61.05 | |
| StateKVBackbone=QWen3-VL-2B, cache budget B=40962026.05 | 60.93 | |
| Full SABackbone=QWen3-VL-2B2026.05 | 60.45 | |
| Full SABackbone=InternVL3-2B2026.05 | 60.22 | |
| ReKVBackbone=InternVL3-8B, retrieval window R=162026.05 | 56.03 | |
| Full SABackbone=InternVL3-1B2026.05 | 55.79 | |
| StateKVBackbone=InternVL3-1B, cache budget B=40962026.05 | 55.44 | |
| ReKVBackbone=Eagle2.5-8B, retrieval window R=162026.05 | 55.44 | |
| ReKVBackbone=QWen3-VL-8B, retrieval window R=162026.05 | 53.88 | |
| ReKVBackbone=QWen3-VL-4B, retrieval window R=162026.05 | 49.22 | |
| ReKVBackbone=QWen3-VL-2B, retrieval window R=162026.05 | 46.71 | |
| ReKVBackbone=InternVL3-1B, retrieval window R=162026.05 | 37.75 | |
| ReKVBackbone=InternVL3-2B, retrieval window R=162026.05 | 33.33 |