[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-step-aos-agent-native-os":3,"topics-all":36,"news-related-2bbd5141-7e38-41bb-9a44-4972af73602d":55},{"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},"2bbd5141-7e38-41bb-9a44-4972af73602d","阶跃星辰推出\"全球首个智能体原生 OS\"Step AOS:把 LLM 当 OS 公民,MCP 拆碎系统调用","7 月 13 日,阶跃星辰在上海一口气端出三件新东西:智能体原生系统 Step AOS、终端品牌 STEPX,以及被称为\"全球首款大模型原生智能体手机\"的 STEPX Neo。和\"Android 加一层 AI 套件\"的做法不同,Step AOS 从 Android、Linux、RTOS 三层一路重写到内核,把 LLM 当作 OS 一等公民。\n技术上最有意思的是能力组织方式。Step AOS 把通信、应用、文件、设置按 MCP(Model Context Protocol)标准拆成更小的\"可被智能体直接调用\"单元——相当于把 MCP 从工具调用协议提升为 OS 系统调用协议。内嵌智能体 Amoo 用\"双域三步记忆\":用户域记住习惯、智能体域沉淀经验,官方称召回可压到 15 毫秒,LongMemEval 排名第一;底层端侧模型是 Step Edge,据称在 29 项基准上排第一,但对标对象未披露。\n执行链走\"端云多脑\"——根据复杂度、成本、延迟动态选 on-device 还是 cloud 模型,首批生态合作方包括 WPS、美团、剪映、携程、高德、支付宝、百度、京东、美图秀秀、微博等。安全方面,阶跃提出\"可信执行环境+每步可审计+按需授权+一键撤回\"四维框架,并与上海人工智能实验室联合发布智能体安全白皮书。STEPX Neo 还通过了目前开放等级最高的国家\"AI 终端智能化分级\"L3 认证,是首款拿到该证书的手机。\n评论:这条路径的真正信号,是大模型公司正绕过 App 层、以 OS 一等公民身份收编终端体验——和 Humane AI Pin、Rabbit r1 这类\"独立 AI 硬件\"路线相反,也是对 OpenAI 传闻中\"AI 手机\"的一次中国式抢答。值得保留怀疑:端侧 29 项第一没公开对标清单、生态对接仍以调用既有 App API 为主、L3 认证的评测细则尚未公开——演示之外,落地到用户手里的体验还有一段路要走。","https:\u002F\u002Fnews.qq.com\u002Frain\u002Fa\u002F20260713A0AW2900","5f7d17cd-f95b-4a76-be2e-db79144de285",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"6ad31a14-c0da-42df-81fd-564281f768db","agentic-ai",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"a8002d98-9df1-4ab9-94d4-a7625af634c4","china-ai",{"id":18,"name":19,"slug":19,"description":13,"color":13},"01598627-1ea6-4b27-a5d8-874971571a71","llm",{"id":21,"name":22,"slug":22,"description":13,"color":13},"b1853a5a-d940-42b7-94f9-0488ee3f2cf7","new-model",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":13},"91885e44-b3d8-4b74-acc2-c632a35c42ed","en","StepFun's Step AOS: an agent-native OS with LLMs as citizens","On July 13, Stepfun unveiled three new things at once in Shanghai: the Agent-native system Step AOS, the terminal brand STEPX, and the STEPX Neo, dubbed \"the world's first large-model-native Agent phone\". Different from the \"Android + a layer of AI kit\" approach, Step AOS rewrites from Android, Linux, and RTOS all the way down to the kernel, treating the LLM as a first-class OS citizen. The most interesting technical piece is how capabilities are organized. Step AOS breaks communication, apps, files, and settings into smaller \"directly callable by Agents\" units according to the MCP (Model Context Protocol) standard — equivalent to elevating MCP from a tool-call protocol to an OS system-call protocol. The embedded Agent Amoo uses a \"dual-domain three-step memory\": the user domain remembers habits, the Agent domain accumulates experience, with the official claim that recall can be compressed to 15 ms and LongMemEval ranks first; the underlying on-device model is Step Edge, claimed to be first across 29 benchmarks — but the comparator wasn't disclosed. The execution chain goes \"end-cloud multi-brain\" — dynamically selecting on-device or cloud models based on complexity, cost, and latency; the first batch of ecosystem partners includes WPS, Meituan, CapCut, Ctrip, Amap, Alipay, Baidu, JD, Meitu Xiuxiu, and Weibo. On the security side, Stepfun proposes a \"trusted execution environment + per-step auditability + on-demand authorization + one-click revocation\" four-dimension framework, and jointly releases an Agent-security white paper with Shanghai AI Lab. STEPX Neo has also passed the currently highest-open-level national \"AI Terminal Intelligence Grading\" L3 certification — the first phone to receive this certificate. Commentary: the real signal of this path is that LLM companies are bypassing the App layer and incorporating the terminal experience as first-class OS citizens — the opposite of the \"standalone AI hardware\" route of Humane AI Pin and Rabbit r1, and a Chinese response to the rumored OpenAI \"AI phone\". Skepticism worth keeping: the on-device 29 firsts don't have a public comparator list, the ecosystem connections are still mostly calling existing App APIs, and the L3 certification's evaluation details haven't been made public — beyond the demo, the experience actually delivered to users still has some distance to go.","step-aos-agent-native-os","2026-07-14T02:30:00Z","2026-07-13T22:04:20.293759Z","2026-08-19T02:08:40.142862Z",true,"agent",231,[37,46],{"slug":38,"tag_slug":38,"title_zh":39,"title_en":40,"intro_zh":41,"intro_en":42,"id":43,"is_active":33,"created_at":44,"modified_at":45},"ai-for-science","AI for Science 2026：从 UniPert 到 GPT-Rosalind 的硬核进化","AI for Science 2026: from UniPert to GPT-Rosalind","生命科学、化学材料、物理世界模型——AI 正在从\"语言工具\"变成\"实验伙伴\"。本专题收录 AI 在三大科学方向的关键节点：UniPert 统一基因与化学扰动空间、GPT-Rosalind 端到端生命科学推理、达摩院 AI 智能体 28 小时找到 4 种超导新材料、Anthropic Claude Science 把工作台做成标准品。","From language tool to lab partner — AI is reshaping life sciences, chemistry\u002Fmaterials, and physical world models. This topic covers the key milestones: UniPert unifying genetic-chemical perturbation spaces, GPT-Rosalind's end-to-end life-sciences reasoning, DAMO's AI agent discovering 4 superconducting materials in 28 hours, and Anthropic's Claude Science workbench going mainstream.","988a4300-5fab-41c4-b5d8-63711a2dc757","2026-09-10T01:34:15.296649Z","2026-09-10T01:34:15.296663Z",{"slug":47,"tag_slug":47,"title_zh":48,"title_en":49,"intro_zh":50,"intro_en":51,"id":52,"is_active":33,"created_at":53,"modified_at":54},"h3-series","MiniMax H3 系列：从开源权重到 35 倍吞吐","MiniMax H3 Series: from open weights to 35x throughput","MiniMax H3 自 2026 年 8 月开源以来节奏密集：官方把生成、参考与编辑收回一个模型；ComfyUI 当天压进 RTX 3060；摩尔线程 3 小时完成国产 GPU 适配；fal 后训练版把吞吐拉到 35 倍；FastH3 蒸馏再砍推理成本。本专题持续追踪 H3 的发布—开源—蒸馏—部署全链路。","Since MiniMax open-sourced H3 in August 2026 the pace has been relentless: one unified omni-modal model, same-day ComfyUI support down to an RTX 3060, a 3-hour Day-0 port to Moore Threads GPUs, fal's post-trained H3 Max at 35x throughput, and FastH3 distillation cutting inference cost further. This topic tracks the full H3 chain — release, open weights, distillation, deployment.","83ef0daa-3c31-4cb3-86ed-e5ee58654d5f","2026-09-08T07:33:19.942193Z","2026-09-08T07:33:19.942209Z",{"items":56},[57,62,67,72,77,82],{"id":58,"title":59,"news_slug":60,"published_at":61},"44aa8908-1cef-48d6-b224-7de12a8d4afd","NeoHorse-1：让 Agent 执行轨迹进入自我改进回路","neohorse-1-agentic-post-training-rsi","2026-09-09T07:19:25+00:00",{"id":63,"title":64,"news_slug":65,"published_at":66},"9822a1a7-0014-4bd5-bbe0-492401fe6b96","AllSpark 把搜索 Agent 推到 BrowseComp 88.6:SFT-RL Climbing 与推理时上下文管理","allspark-iris-search-agent-sft-rl-climbing","2026-09-07T07:11:17+00:00",{"id":68,"title":69,"news_slug":70,"published_at":71},"2fc64783-8b2a-49a3-939b-edf02bff3622","Ox Alpha 指纹指向 GLM-5.3:OpenRouter 的 1M 上下文隐身模型可能是智谱","ox-alpha-glm-5-3-stealth-zhipu","2026-08-22T14:00:00+00:00",{"id":73,"title":74,"news_slug":75,"published_at":76},"d1e8997e-bb60-453d-9ef8-71b8bdde5386","Harvey 首个自研法律模型 Tenet 曝光:底座没选 GPT 和 Claude,选了 Kimi K3","harvey-tenet-kimi-k3-legal-model","2026-08-18T17:30:00+00:00",{"id":78,"title":79,"news_slug":80,"published_at":81},"8bd5a96a-b85b-4db6-ad54-a2c311867178","字节跳动被曝训练10万亿参数超大模型：对标Anthropic Mythos,中国LLM进入\"10T俱乐部\"前夜","bytedance-10-trillion-parameter-model","2026-08-07T09:11:00+00:00",{"id":83,"title":84,"news_slug":85,"published_at":86},"2ada2e69-25c2-4951-9e49-2b24a043393e","腾讯 Marvis 把 Agent 拽到端侧:混元要做 PC 集群,应用宝做了「系统级」分诊","tencent-marvis-on-device-agent","2026-07-23T20:30:00+00:00"]