[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-nvidia-cosmos-3-edge-4b":3,"news-related-b446777f-9949-47e4-8ca8-2d0fa7f14126":36},{"id":4,"title":5,"summary":6,"content":6,"original_url":7,"source_id":8,"tags":9,"translations":23,"news_slug":29,"published_at":30,"created_at":31,"modified_at":32,"is_published":33,"publish_type":34,"image_url":13,"view_count":35},"b446777f-9949-47e4-8ca8-2d0fa7f14126","NVIDIA Cosmos 3 Edge 4B开源世界模型：物理AI实时推理从云端搬到产线","2026年7月16日，NVIDIA正式推出Cosmos 3 Edge——4B参数世界模型，专为机器人和视觉智能体在边缘端实时推理而生，补全Cosmos 3家族。\n\n和Super\u002FNano走\"数据中心+合成数据\"路线不同，Cosmos 3 Edge把模型搬到设备侧运行，目标是Jetson Thor这类边缘算力平台。工厂机械臂不能等往返数据中心的RTT判断\"那个移动的影子是不是工人\"，必须在毫秒级内决策。\n\nCosmos 3 Edge基于NVIDIA Nemotron底座构建，开发者可借助开源Cosmos框架针对特定机器人、传感器做后训练——NVIDIA称\"约一天完成embodiment适配\"。\n\nCosmos Coalition在日本同步扩容，FANUC、川崎重工、索尼、安川电机、软银、NEC等十数家工业巨头加入，目标是把机器人、自动化系统的推理负载下放到设备端。\n\n4B意味着Cosmos 3 Edge不依赖云端，但也不完全脱离——更可能是混合栈：云端训练、合成数据、仿真，边缘负责实时感知与执行。物理AI的下一程，已从\"能不能跑\"切换到\"能不能在产线上跑\"。","https:\u002F\u002Fnvidianews.nvidia.com\u002Fnews\u002Fjapans-robotics-and-manufacturing-leaders-build-on-nvidia-cosmos-to-advance-physical-ai-frontier","474eef8c-e0c3-46cf-adee-c089558220f9",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"7e89b5cc-57db-4f37-bc6d-28919a73931c","model-release",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"8dac812d-3839-4abe-a855-5f56ec9515fd","nvidia",{"id":18,"name":19,"slug":19,"description":13,"color":13},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",{"id":21,"name":22,"slug":22,"description":13,"color":13},"ebe5dcd1-46b1-4298-b8c2-8e0e2f456e56","video-generation",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":13},"961ee24e-e33f-4c58-961c-dad153a51fb1","en","NVIDIA Cosmos 3 Edge 4B: world model moves to the factory line","On July 16, 2026, NVIDIA officially launched Cosmos 3 Edge — a 4B-parameter world model, purpose-built for real-time inference on edge devices by robots and visual agents, completing the Cosmos 3 family. Unlike Super\u002FNano, which follow a \"data center + synthetic data\" path, Cosmos 3 Edge moves the model onto the device side, targeting edge-compute platforms like Jetson Thor. A factory robotic arm can't wait for a round-trip-to-data-center RTT to decide \"is that moving shadow a worker?\" — it has to decide in milliseconds. Cosmos 3 Edge is built on the NVIDIA Nemotron base, and developers can use the open-source Cosmos framework to post-train for specific robots and sensors — NVIDIA claims \"embodiment adaptation in about a day\". The Cosmos Coalition simultaneously expanded in Japan, with FANUC, Kawasaki Heavy Industries, Sony, Yaskawa, SoftBank, NEC and a dozen other industrial giants joining, with the goal of moving robotics and automation-system inference workloads down to the device side. 4B means Cosmos 3 Edge doesn't depend on the cloud, but doesn't entirely leave it either — the more likely architecture is a hybrid stack: cloud-side training, synthetic data, and simulation, with the edge handling real-time perception and execution. The next mile of physical AI has already shifted from \"can it run\" to \"can it run on the production line\".","nvidia-cosmos-3-edge-4b","2026-07-16T14:00:00Z","2026-07-16T14:09:29.895212Z","2026-08-19T02:08:40.142862Z",true,"agent",295,{"items":37},[38,43,48,53,58,63],{"id":39,"title":40,"news_slug":41,"published_at":42},"5612d186-46ee-4509-9a93-94045ba004ae","LTX-2.5 开放权重视频模型:4K 反而在 Fast 端点,EXR 色彩管线也焊进去了","ltx-2-5-open-weights-video","2026-08-18T15:20:00+00:00",{"id":44,"title":45,"news_slug":46,"published_at":47},"de2cceb2-7d39-4a5f-844e-5a3144667f49","Nemotron 3.5 Lightning 开源：30B 总参 3B 激活的混合 MoE，直接用 NVFP4 配方预训练","nemotron-35-lightning-30b-a3b-open-release","2026-08-16T15:00:00+00:00",{"id":49,"title":50,"news_slug":51,"published_at":52},"6f9e9f94-9dcc-4c6c-b254-6c5d0fe8ed37","京东开源 JoyAI-Video-Edit:16B 多模态扩散 Transformer 把视频编辑推进「边播边改」实时流时代","jd-joyai-video-edit-realtime-diffusion","2026-08-10T00:00:00+00:00",{"id":54,"title":55,"news_slug":56,"published_at":57},"bcdc10bc-2f08-4c39-8ffa-e7e34041c112","京东开源 JoyAI-Video-Edit:用 16B 多模态扩散 Transformer 把视频编辑推进「边播边改」实时流时代","jd-joyai-video-edit-real-time-streaming","2026-08-05T03:00:00+00:00",{"id":59,"title":60,"news_slug":61,"published_at":62},"6e3002da-c1fd-4a6d-b903-4f65b976dd04","MiniMax H3 首个商用落点：美图 RoboNeo 接入背后,通用多模态模型的\"可编辑性\"才刚开始被检验","roboneo-minimax-h3-multimodal-editing","2026-08-03T18:02:02+00:00",{"id":64,"title":65,"news_slug":66,"published_at":67},"62ca04a1-d7fb-4169-8e31-4724405dd3b4","原力灵机 DW0.5 把强化学习搬进虚拟世界,真机数据需求骤降 60%","yuanli-lingji-dw-0-5-rl-world","2026-07-18T08:01:00+00:00"]