Streaming video understanding on OVOBench (Specific Accuracy Breakdown)
69.1Accuracy (Proactive Forwarding)ReKV
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| ReKVSize=7B, Offload=False2026.03 | 69.1 | — | — | — | |
| StreamReadySize=7B, Backbone=Qwen-2-VL2026.03 | 58.8 | 63.7 | 72.2 | 68.2 | |
| ViSpeakSize=7B, Backbone=Qwen-2-VL2026.03 | 54.3 | 61.6 | 57.5 | 59.4 | |
| LLaVA-NeXT-VideoSize=7B2026.03 | 54.2 | 41.7 | 41.7 | 53.1 | |
| VITA 1.5Size=7B2026.03 | 53.5 | 33.1 | 41.5 | 52.8 | |
| StreamForestSize=7B, Backbone=Qwen-2-VL2026.03 | 53.5 | — | 52 | 55.6 | |
| LLaVA-OneVisionSize=7B2026.03 | 50.9 | 38.4 | 45 | 52.9 | |
| Qwen-2-VLSize=7B2026.03 | 48.9 | 34.9 | 48.6 | 52.7 | |
| LongVUSize=7B2026.03 | 48.5 | — | 39.5 | 48.5 | |
| StreamBridgeSize=7B, Backbone=Qwen-2-VL2026.03 | 48.4 | 24.1 | 68.1 | 62.6 | |
| HierarQSize=7B2026.03 | 48.3 | 44.4 | 48.3 | 54.6 | |
| InfiniPot-VSize=7B, Backbone=Qwen-2-VL2026.03 | 47.9 | — | 47.6 | 53.8 | |
| InternVL2Size=8B2026.03 | 45.4 | 35.8 | 44 | 50 | |
| StreamAgentSize=7B, Backbone=Qwen-2-VL2026.03 | 45.4 | 36.3 | 41.7 | 49.5 | |
| Flash-VStreamSize=7B, Backbone=Qwen-2-VL2026.03 | 44.2 | 26 | 25.4 | 33.2 | |
| TimeChat-OnlineSize=7B2026.03 | 36.4 | 37.8 | 42 | 45.7 | |
| DispiderSize=7B, Backbone=Qwen-2-VL2026.03 | 34.7 | 35.7 | 36.1 | 41.8 | |
| LongVASize=7B2026.03 | — | 35.9 | — | — | |
| MiniCPM-V 2.6Size=8B2026.03 | — | 35 | — | — | |
| VideoLLM-onlineSize=8B2026.03 | — | 28.5 | 17.7 | 19.3 |