We introduce Light-Omni, a highly efficient multimodal agent framework for long-term video understanding. Advanced video agents usually suffer from prohibitive latency due to heavy \"detective-style\" iterative reasoning. To solve this, Light-Omni enables instant \"reflexive\" responses via a novel dual-state mechanism (Global &amp; Latent states) in a single forward pass.</p>\n<p>Key Highlights:</p>\n<ul>\n<li>🔥 Incredible Efficiency: Achieves a 12.1× speedup and 2.6× GPU memory reduction compared to M3-Agent, with near-constant latency (~2.3s) regardless of video length.</li>\n<li>📈 SOTA Performance: Delivers an average accuracy of 64.8% across VideoMME-long, LVBench, and HippoVlog.</li>\n<li>🛠️ Plug-and-Play: Acts as a foundational memory system that seamlessly boosts existing MLLMs (e.g., Qwen2.5-VL, Qwen3-VL, Gemini-2.0-Flash).</li>\n</ul>\n<p>Code, models, and dataset are available in our repo!</p>\n","updatedAt":"2026-07-08T01:53:54.314Z","author":{"_id":"69b14cdeeed94a88011ce095","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/1LRe2njtqTGIsBTkOSIOT.jpeg","fullname":"Nie","name":"ClareNie","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":2,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8467066884040833},"editors":["ClareNie"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/1LRe2njtqTGIsBTkOSIOT.jpeg"],"reactions":[],"isReport":false}},{"id":"6a4db01c7bf60b5e9d6e21d7","author":{"_id":"69e6cb8f0e62a5acf730ab0c","avatarUrl":"/avatars/a406261d33385ebe6fc90e9754949e04.svg","fullname":"Kelly Wak","name":"2924clare","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false},"createdAt":"2026-07-08T02:04:12.000Z","type":"comment","data":{"edited":true,"hidden":true,"hiddenBy":"","latest":{"raw":"This comment has been hidden","html":"This comment has been hidden","updatedAt":"2026-07-08T02:07:29.483Z","author":{"_id":"69e6cb8f0e62a5acf730ab0c","avatarUrl":"/avatars/a406261d33385ebe6fc90e9754949e04.svg","fullname":"Kelly Wak","name":"2924clare","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":1,"editors":[],"editorAvatarUrls":[],"reactions":[]}},{"id":"6a4dde9f212c7bb59cd4c40e","author":{"_id":"69b14cdeeed94a88011ce095","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/1LRe2njtqTGIsBTkOSIOT.jpeg","fullname":"Nie","name":"ClareNie","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":2,"isUserFollowing":false},"createdAt":"2026-07-08T05:22:39.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"paper: https://arxiv.org/abs/2607.05511\ncode: https://github.com/Clare-Nie/Light-Omni\nwebsite: https://clare-nie.github.io/Light-Omni/","html":"<p>paper: <a href=\"https://arxiv.org/abs/2607.05511\" rel=\"nofollow\">https://arxiv.org/abs/2607.05511</a><br>code: <a href=\"https://github.com/Clare-Nie/Light-Omni\" rel=\"nofollow\">https://github.com/Clare-Nie/Light-Omni</a><br>website: <a href=\"https://clare-nie.github.io/Light-Omni/\" rel=\"nofollow\">https://clare-nie.github.io/Light-Omni/</a></p>\n","updatedAt":"2026-07-08T05:22:39.856Z","author":{"_id":"69b14cdeeed94a88011ce095","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/1LRe2njtqTGIsBTkOSIOT.jpeg","fullname":"Nie","name":"ClareNie","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":2,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.5483754277229309},"editors":["ClareNie"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/1LRe2njtqTGIsBTkOSIOT.jpeg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2607.05511","authors":[{"_id":"6a4da52425849b193a8349fd","name":"Chang Nie","hidden":false},{"_id":"6a4da52425849b193a8349fe","name":"Jiaju Wei","hidden":false},{"_id":"6a4da52425849b193a8349ff","name":"Junlan Feng","hidden":false},{"_id":"6a4da52425849b193a834a00","name":"Chaoyou Fu","hidden":false},{"_id":"6a4da52425849b193a834a01","name":"Caifeng Shan","hidden":false}],"publishedAt":"2026-07-06T00:00:00.000Z","submittedOnDailyAt":"2026-07-08T00:00:00.000Z","title":"Light-Omni: Reflex over Reasoning in Agentic Video Understanding with Long-Term Memory","submittedOnDailyBy":{"_id":"69b14cdeeed94a88011ce095","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/1LRe2njtqTGIsBTkOSIOT.jpeg","isPro":false,"fullname":"Nie","user":"ClareNie","type":"user","name":"ClareNie"},"summary":"Agentic video understanding equips models with long-term memory to autonomously process and respond to continuous, long-horizon multimodal streams. However, advanced video agents often rely on ``detective-style'' iterative reasoning for action control (e.g., search) and evidence aggregation, incurring prohibitive costs and latency. We argue that such heavy reasoning primarily compensates for the lack of global context and semantic misalignment in retrieval. This paper introduces Light-Omni, a multimodal agent framework for reflexive and lightweight video understanding. It achieves this through dual contextual states that instantly build the required context in a single forward pass. First, we maintain a global state, a finite-sized multimodal script continuously consolidated from episodic memory, serving as the global context for Light-Omni. Through hierarchical merging, it preserves recent details while summarizing past events. Second, conditioned on this global context, we generate a parametric latent state that directly drives autonomous actions and produces retrieval embeddings, with minimal latency. Benefiting from this coupled design, Light-Omni achieves semantically aligned retrieval and reflexive responses while avoiding iterative reasoning. Extensive experiments validate the effectiveness of Light-Omni across multiple video benchmarks. Notably, it outperforms M3-Agent with an average 2.4% accuracy gain, a 12.1times speedup, and a 2.6times improvement in GPU memory efficiency. Furthermore, it serves as a memory system to enhance both the performance and efficiency of existing MLLMs. Project page: https://clare-nie.github.io/Light-Omni.","upvotes":19,"discussionId":"6a4da52525849b193a834a02","projectPage":"https://clare-nie.github.io/Light-Omni","githubRepo":"https://github.com/Clare-Nie/Light-Omni","githubRepoAddedBy":"user","ai_summary":"Light-Omni is a multimodal agent framework that enables efficient video understanding through dual contextual states, achieving faster and more accurate video processing by eliminating iterative reasoning while maintaining semantic alignment.","ai_keywords":["multimodal agent framework","reflexive video understanding","global state","episodic memory","parametric latent state","retrieval embeddings","iterative reasoning","semantic alignment","hierarchical merging","video benchmarks","MLLMs"],"ai_summary_model":"Qwen/Qwen2.5-Coder-32B-Instruct","githubStars":25,"organization":{"_id":"638f70e8f1256a80d4288555","name":"nanjinguniv","fullname":"Nanjing University","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/638f706ef1256a80d42880f9/6M6-JzwJGiLxjIJzvCflf.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"69b14cdeeed94a88011ce095","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/1LRe2njtqTGIsBTkOSIOT.jpeg","isPro":false,"fullname":"Nie","user":"ClareNie","type":"user"},{"_id":"69e62c0be8d236a10f608f69","avatarUrl":"/avatars/c5de9a26daf4b3bac95e4878c9bf16a8.svg","isPro":false,"fullname":"Kelly Wakasa","user":"Wangyizz","type":"user"},{"_id":"6380993fbb68e3bef8ccdf29","avatarUrl":"/avatars/9b6a002207796c0aa5f10fa8e6d56296.svg","isPro":false,"fullname":"chang","user":"clare3","type":"user"},{"_id":"69e6104e87b7b13ee642b5d1","avatarUrl":"/avatars/c9115b98e90684211f11f1ad2566021a.svg","isPro":false,"fullname":"Kelly Wakasa","user":"JunfangF","type":"user"},{"_id":"69e6cb8f0e62a5acf730ab0c","avatarUrl":"/avatars/a406261d33385ebe6fc90e9754949e04.svg","isPro":false,"fullname":"Kelly Wak","user":"2924clare","type":"user"},{"_id":"663ad94fa55b0634631b4712","avatarUrl":"/avatars/29285d9aeaab23e080304144a8a0d1bd.svg","isPro":false,"fullname":"huggingyan","user":"huggingyan","type":"user"},{"_id":"6a18f8b1af364fc647a6e3e2","avatarUrl":"/avatars/b14adb25ecb9bbe2f0dbef7633a04981.svg","isPro":false,"fullname":"Qingyuan Cao","user":"qingyuannju","type":"user"},{"_id":"6a2da6c8ca070ee12c6e396c","avatarUrl":"/avatars/0355287dcabaa67dbc7f0b10b87451f9.svg","isPro":false,"fullname":"Joe Mama","user":"JoeMama123123123","type":"user"},{"_id":"6555766238df282b0f0454bd","avatarUrl":"/avatars/eab271dee045e0aaf8a65fc3fe9acafe.svg","isPro":false,"fullname":"Shaoqi Dong","user":"ShaoqiDong","type":"user"},{"_id":"678f682784026ea238b0786d","avatarUrl":"/avatars/f7ca68f366386d13373fac5118a00b5d.svg","isPro":false,"fullname":"牛永辉","user":"summit8848","type":"user"},{"_id":"68db9b2dadc11a1b70408123","avatarUrl":"/avatars/428b4ba13ee57c598a786d9421e68a10.svg","isPro":true,"fullname":"Ruoliu Yang","user":"RuoliuYang","type":"user"},{"_id":"67da446beb707e7f71d78d05","avatarUrl":"/avatars/9385bb28f35af7ff6e8473bc20e6d3b9.svg","isPro":false,"fullname":"Flyoung","user":"Flyoung","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"638f70e8f1256a80d4288555","name":"nanjinguniv","fullname":"Nanjing University","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/638f706ef1256a80d42880f9/6M6-JzwJGiLxjIJzvCflf.png"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2607/2607.05511.md","query":{}}">

Get this paper in your agent:

Don't have the latest CLI?

Add this paper to a collection to link it from this page.