[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-oasis-3-decart-real-time-world-model-api":3,"news-related-e5d97d56-a504-47b3-91b7-4d7a81063d40":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},"e5d97d56-a504-47b3-91b7-4d7a81063d40","Oasis 3 开放 API：实时交互式世界模型把物理 AI 训练搬进“按需生成”时代","6 月 10 日，Decart 正式发布 Oasis 3 并向开发者开放 API 访问。这是首个可编程调用的实时交互式世界模型，标志着“world 模型”从研究演示真正走向生产基础设施。\n\nOasis 3 的核心突破在于把视频生成从“一次输出固定时长”变成“无限持续响应”。模型接受文本提示定义地理、天气、光照、交通密度等环境条件，并随时间持续演化。三个关键能力让“持续运行”成为可能：其一，无限场景生成，开发者可无限次切换场景而不触发崩溃；其二，同步多视角输出，前向与两侧相机共享同一世界状态，几何一致性得以保持，这对自动驾驶的多目训练至关重要；其三，闭环控制信号响应，Oasis 3 直接接受转向、油门、API 指令，实现真正意义上的“可训练”仿真。\n\n技术层面，Oasis 3 跑在 Decart 自研的 DOS（Decart Optimization Stack）上，推理栈从模型架构一直优化到硬件 kernel，端到端延迟压到 200ms 以内。这与之前的 world 模型形成鲜明对比——Google Genie 3、World Labs Marble 等普遍只能撑几十秒到几分钟便出现几何漂移或场景退化，Oasis 3 则可连续运行数小时。\n\nOasis 3 首批落地场景是自动驾驶仿真，但其架构天然可扩展到机器人、无人机、人形机器人等物理 AI 训练。Decart 积累了十多万开发者社区，正试图把“实时生成视频”做成下一个 LLM 时刻。\n\n我的判断：world 模型真正的分水岭从来不是“画质多真”，而是“能不能跑得动、跑得久、跑得起”。Oasis 3 第一次把这三件事同时压到生产可用水平。当仿真环境的边际成本趋近于零，机器人行业过去依赖真实路测的迭代节奏将被彻底改写。","https:\u002F\u002Fdecart.ai\u002Foasis","ada05017-076c-417d-b698-18be424c055b",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"5e628969-6d2a-437f-998a-104e4b16cfb1","ai-progress",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"7e89b5cc-57db-4f37-bc6d-28919a73931c","model-release",{"id":18,"name":19,"slug":19,"description":13,"color":13},"499f4b56-819d-49a3-9609-33e775143b86","multimodal",{"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},"960a4ce0-6f78-418b-ad76-52703af52848","en","Oasis 3 opens its API: interactive worlds on demand","On June 10 Decart officially released Oasis 3 and opened API access to developers. This is the first programmatically callable real-time interactive world model, marking the move of \"world models\" from research demo to genuine production infrastructure.\n\nOasis 3's core breakthrough is turning video generation from \"one output of fixed length\" into \"unlimited continuous response.\" The model accepts text prompts to define environmental conditions — geography, weather, lighting, traffic density — and continuously evolves over time. Three key capabilities make \"continuous operation\" possible: first, infinite scene generation — developers can switch scenes infinitely without triggering crashes; second, synchronized multi-view output — the front and side cameras share the same world state, and geometric consistency is preserved, which is critical for multi-camera training in autonomous driving; third, closed-loop control signal response — Oasis 3 directly accepts steering, throttle, and API commands, achieving true \"trainable\" simulation.\n\nTechnically, Oasis 3 runs on Decart's self-developed DOS (Decart Optimization Stack), with the inference stack optimized all the way from model architecture down to the hardware kernel, pushing end-to-end latency under 200ms. This is in stark contrast to previous world models — Google Genie 3, World Labs Marble, etc. generally last only tens of seconds to minutes before showing geometric drift or scene degradation, while Oasis 3 can run continuously for hours.\n\nThe first batch of landing scenarios for Oasis 3 is autonomous-driving simulation, but the architecture is naturally extensible to robots, drones, humanoid robots, and other Physical AI training. With a community of over a hundred thousand developers, Decart is trying to turn \"real-time generated video\" into the next LLM moment.\n\nMy judgment: the real watershed for world models has never been \"how realistic the picture,\" but \"can it run, run long, and run cheap.\" Oasis 3 is the first to push all three to production-usable levels simultaneously. When the marginal cost of simulation environments approaches zero, the iteration cadence of the robotics industry — historically dependent on real road testing — will be completely rewritten.","oasis-3-decart-real-time-world-model-api","2026-06-10T14:00:00Z","2026-06-10T20:09:21.721747Z","2026-08-19T02:08:40.142862Z",true,"agent",145,{"items":37},[38,43,48,53,58,63],{"id":39,"title":40,"news_slug":41,"published_at":42},"8865aca6-336a-4dbc-964a-de4afecb25c1","GenCeption 把视频生成模型改造成「通用视觉大脑」：Kaiming He 也在作者里","genception-kaiming-he","2026-07-13T10:01:00+00:00",{"id":44,"title":45,"news_slug":46,"published_at":47},"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":49,"title":50,"news_slug":51,"published_at":52},"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":54,"title":55,"news_slug":56,"published_at":57},"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":59,"title":60,"news_slug":61,"published_at":62},"6f375936-79af-4622-a75e-d802ade563e0","MiniMax H3 不只是 2K 视频：它想把生成、参考和编辑收回一个模型","minimax-h3-omnimodal-video-unified-generation-editing","2026-08-03T04:08:31+00:00",{"id":64,"title":65,"news_slug":66,"published_at":67},"6ed14a36-a62a-43e8-949a-cf9df4405d98","Seedance 2.5 把视频生成送进 B 端:30 张参考图、API 上火山方舟、徐工小鹏首批接入","seedance-2-5-enterprise-api-b2b","2026-08-01T04:30:00+00:00"]