Native Active Perception as Reasoning for Omni-Modal Understanding

Passive models for long video understanding typically rely on a "watch-it-all" paradigm, processing frames uniformly regardless of query difficulty, causing computational cost to grow with video duration. Although interactive frameworks have emerged, they often rely on global pre-scanning, and their context cost still scales with video length. We propose **OmniAgent**, the first native omni-modal agent that formulates video understanding as a POMDP-based iterative **Observation-Thought-Action** cycle. OmniAgent executes on-demand actions to selectively distill audio-visual cues into a persistent textual memory, effectively decoupling reasoning complexity from raw video duration. To operationalize this, we introduce (1) **Agentic Supervised Fine-Tuning** to bootstrap native active perception via best-of-N trajectory synthesis with dual-stage quality control, and (2) **Agentic Reinforcement Learning** with **TAURA** (Turn-aware Adaptive Uncertainty Rescaled Advantage), which leverages turn-level entropy to steer credit assignment toward pivotal discovery turns. Crucially, OmniAgent exhibits positive test-time scaling, where performance improves as the number of reasoning turns increases, validating the efficacy of active perception. Empirical results across ten benchmarks (e.g., VideoMME, LVBench) demonstrate that OmniAgent achieves state-of-the-art performance among open-source models. Notably, on LVBench, our 7B agent outperforms the $10\times$ larger Qwen2.5-VL-72B (50.5% vs. 47.3%). We release our code and model at https://github.com/HarryHsing/OmniAgent.

When AI systems need to answer a question about a video — say, finding who appears in the opening scene of a two-hour film — they typically process the entire video from start to finish. For long videos, this demands enormous computational resources, and by treating every moment equally, these systems can overlook key details hidden in a brief instant. We present OmniAgent, an AI agent that watches videos the way people do: instead of processing everything, it actively decides where to look and what to listen for based on the question asked. Step by step, it skims through clips, listens to relevant audio, and takes notes until it gathers enough evidence to answer — skipping irrelevant content. This active approach makes video understanding far more efficient. Our agent — despite being relatively small — outperforms AI models ten times its size while examining far fewer video frames. By making it practical for AI to reason over hour-long videos without watching every second, this work can improve applications such as video search, media accessibility, and content analysis.