[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-tencent-marvis-on-device-agent":3,"topics-all":36,"news-related-2ada2e69-25c2-4951-9e49-2b24a043393e":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},"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","WAIC 上,腾讯副总裁林松涛给了一组数字:**Marvis 上线两天 DAU 突破 30 万,七日留存约 54%。** 这条线刻意避开主流——不卷通用 Agent,不抢豆包、元宝的文档问答市场,只死磕「系统级操作」。\n\nMarvis 负责人蔡建涛披露的用户场景分布很说明问题:**本地文件 44%,电脑硬件管理 28%,浏览器任务 18%,应用相关 16%,搜索类只有 6%。** 跟主流 Agent 以「搜索 + 文档总结」为主的画像完全不同——用户拿 Marvis 主要在「管电脑」,而不是「找资料」。\n\n团队明确表态**不做手机端侧,理由很技术:「手机内存和电池有限,放不下能跑复杂任务的模型」**。他们把赌注下到 PC、Mini PC、AI Box 这类比手机算力大一截、又比云端服务器轻的设备上,手机只做控制端,大算力留在桌面盒子或云端容器里。\n\n林松涛还透露,**Marvis 正和混元一起推出面向 PC 端的端侧模型集群**。这跟混元「撤多模态理解、押世界模型」的调整看似矛盾,实际是端云分工的细化:小模型装进盒子做本地文件、上下文;大模型留在云端处理重推理。\n\n行业层面值得关注三点:**一是**端侧模型集群的工程难点不是「塞进去」而是跨设备状态一致性——手机端状态怎么同步给桌面盒子继续推理,是 2026 年 Agent 落地最难的拼图;**二是**国内做端侧大模型的玩家都在卷「纯终端」,腾讯这条「本地 Agent + 云端容器」的混合架构走出了第三条路;**三是**当 DAU 30 万、七日留存 54% 这种数字被公布后,行业得重新评估「Agent 必须能搜资料写文档」这个范式是不是被过度营销了。","https:\u002F\u002F36kr.com\u002Fp\u002F3907676111983745","5e4fd3d1-9cb4-44a6-bae5-9ffb449c05c1",[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},"e676a5cf-1f24-472f-a765-86fa21a1bc3c","ai-model",{"id":18,"name":19,"slug":19,"description":13,"color":13},"a8002d98-9df1-4ab9-94d4-a7625af634c4","china-ai",{"id":21,"name":22,"slug":22,"description":13,"color":13},"01598627-1ea6-4b27-a5d8-874971571a71","llm",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":28},"b1ed6e60-0d10-4dd8-863a-9c9d6ec484e8","en","Tencent Marvis pulls agents on-device with PC clusters","At WAIC, Tencent VP Lin Songtao gave a number: **Marvis hit 300K DAU in its first two days online, with about 54% seven-day retention.** The line deliberately avoids the mainstream — no general-purpose Agent, no battle with Doubao or Yuanbao for document Q&A, just hard grinding on \"system-level operations\". Marvis lead Cai Jiantao's user-scenario distribution says a lot: **local files 44%, hardware management 28%, browser tasks 18%, app-related 16%, search only 6%**. Very different from the mainstream Agent profile of \"search + document summary\" — users come to Marvis mainly to \"manage the computer\", not \"find information\". The team explicitly states **they will not do mobile-device-side, with a technical reason: \"phone memory and battery are limited, can't fit a model that can run complex tasks\"**. They bet on PC, Mini PC, and AI Box — devices with substantially more compute than a phone but lighter than a cloud server. The phone only acts as a controller; the big compute stays in a desktop box or a cloud container. Lin Songtao also revealed that **Marvis is working with Hunyuan to launch a device-side model cluster for PCs**. This seems contradictory to Hunyuan's \"exit multimodal understanding, bet on world model\" adjustment, but is actually a refinement of device-cloud division: small models go in the box for local files and context; big models stay in the cloud for heavy reasoning. Three industry takeaways: **First,** the engineering difficulty of device-side model clusters isn't \"fitting them in\" but cross-device state consistency — how the phone's state syncs to the desktop box to continue inference is the hardest piece of the Agent-deployment puzzle in 2026. **Second,** Chinese on-device large-model players are all racing \"pure-terminal\", and Tencent's \"local Agent + cloud container\" hybrid architecture has carved out a third path. **Third,** when numbers like 300K DAU and 54% seven-day retention get announced, the industry has to reassess whether the \"Agent must search and write documents\" paradigm has been over-marketed.","tencent-marvis-on-device-agent","2026-07-23T20:30:00Z","2026-07-23T20:05:10.346876Z","2026-08-19T02:08:40.142862Z",true,"agent",235,[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},"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":63,"title":64,"news_slug":65,"published_at":66},"ad10985b-425c-4af1-9495-c63792a2b593","腾讯混元把语音识别打到 3% WER：Hy ASR 3.0 preview 让 ASR 从“逐字”走向“读语境”","tencent-hunyuan-hy-asr-3-0-preview-context-aware","2026-08-05T00:00:00+00:00",{"id":68,"title":69,"news_slug":70,"published_at":71},"654e153e-b198-4afb-a236-2e284ca565e4","亚马逊把 Alexa+ 改造成全场景 AI 代理:自然对话+文件读取+日程接管,大模型开始啃消费端","amazon-alexa-plus-ai-agent","2026-07-26T02:30:00+00:00",{"id":73,"title":74,"news_slug":75,"published_at":76},"4d9f11bb-5795-45e5-a90b-7eb29756da24","腾讯混元发布 Hyra-1.0：用递归自我改进重写研究智能体的范式","tencent-hunyuan-hyra-1","2026-07-21T06:30:00+00:00",{"id":78,"title":79,"news_slug":80,"published_at":81},"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","step-aos-agent-native-os","2026-07-14T02:30:00+00:00",{"id":83,"title":84,"news_slug":85,"published_at":86},"eac81652-8158-4b38-a1f5-7ba9420fb74e","清华 AgenticDataBench：把 LLM 数据智能体拉进「真实业务」的统考卷","tsinghua-agenticdatabench","2026-07-03T08:00:00+00:00"]