Video Reasoning on OmniVideoBench
40.52Accuracy (Long)AOP-Agent
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| AOP-AgentBackbone=Qwen3-Omni-Instruct2026.05 | 40.52 | 33.05 | 39.84 | 37.82 | 33.2 | |
| AOP-AgentBackbone=Qwen3-Omni-Thinking2026.05 | 40.52 | 35.59 | 39.84 | 38.66 | 32.9 | |
| Qwen3-Omni-ThinkingProtocol=Direct Inference2026.05 | 31.9 | 37.29 | 37.4 | 35.57 | 31.9 | |
| AOP-AgentBackbone=Qwen2.5-Omni-7B2026.05 | 31.9 | 33.05 | 34.15 | 33.05 | 31.1 | |
| Qwen2.5-Omni-7BProtocol=Direct Inference2026.05 | 30.17 | 27.12 | 34.15 | 30.53 | 29 | |
| Active Video PerceptionFramework Group=Other Video Agent Frameworks, Backbone=Qwen3-Omni-Instruct2026.05 | 30.17 | 22.88 | 36.89 | 30.06 | 26.9 | |
| Qwen3-Omni-InstructProtocol=Direct Inference2026.05 | 28.45 | 26.27 | 38.21 | 31.09 | 29.3 | |
| OmniAgentFramework Group=Other Video Agent Frameworks, Backbone=Qwen3-Omni-Instruct2026.05 | 28.45 | 34.75 | 31.71 | 31.65 | 27.3 |