[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-opengcua-open-computer-use-stack":3,"topics-all":38,"news-related-50687317-89b7-4857-ad66-86e2e251ff9d":57},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":24,"news_slug":31,"published_at":32,"created_at":33,"modified_at":34,"is_published":35,"publish_type":36,"image_url":14,"view_count":37},"50687317-89b7-4857-ad66-86e2e251ff9d","阿里开源 OpenGCUA：8 个前沿模型电脑操控平均涨 19.6 分","阿里 MAI-UI 团队开源电脑操控智能体全栈 OpenGCUA：8 个前沿模型在 OSWorld-v2 平均涨 19.6 分、步数省 53%，真机 harness 与跨设备 SDK 以 Apache-2.0 发布；Qwen3.8-Max 真机推理 token 达 GPT-6 Astra 的 19.4 倍。","GPT-6 Astra 把「AI 操作电脑」推成前沿能力，但那套系统是封闭的——外界只能看 demo，没法拆开研究。阿里 MAI-UI 团队给出了一个对位答案：开源项目 OpenGCUA，目标是把通用电脑操控智能体（Generalist Computer-Use Agent）的全栈开放给社区，首个版本已随技术报告（arXiv:2610.22200，https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.22200）一起放出，代码以 Apache-2.0 许可发布在 GitHub 的 Tongyi-MAI 组织下。\n\n## 不止「截图-点击」：GCUA 换了一种用电脑的方式\n\n团队把 GCUA 定义为一个自己决定「怎么观察、怎么行动、怎么组织工作」的系统：它读无障碍树、调用应用 API、跑 shell 命令，只在必要时才碰 GUI。这对应他们观察到的转向——评价标准从「模型把 GUI 点得多好」变成「系统用一切可行手段把活干完」。\n\n## 两项核心设计\n\n一是代码执行：GUI 操作被写成代码，和 API、shell 调用揉进同一个程序里，循环、分支、中间结果直接驱动下一步。二是 harness-to-model 决策转移：把观察与任务状态决策从固定的 harness 规则挪回模型侧——harness 只提供机制，什么时候要截图、什么时候切工具由模型自己判断。\n\n## 数据：八个前沿模型全部上涨\n\nOSWorld-v2 基准上，GPT-6 Astra、Claude Opus 5、Gemini 3.8 Flash 等 8 个模型平均分从 45.3 涨到 64.9（+19.6），平均步数下降 53%；GPT-6 Astra 单模型步数从 102 降到 34。弱模型受益最大：GLM-5.3-Flash 涨 35.2 分，DeepSeek-V4.1-Flash 涨 29.8 分，最强与最弱模型的差距从 39.1 分缩到 18.0 分。消融还显示，同样给 GUI 和 shell 权限，把操作写成代码比 GUI+CLI 智能体再高 6.1 分、步数再省 14.8%。\n\n## 真机泼的冷水：19.4 倍 token 差距\n\n更有信息量的是真机侧。固定 harness、跑四项 macOS 任务时，Qwen3.8-Max 消耗的推理 token 是 GPT-6 Astra 的 19.4 倍，完成的任务还更少。团队把这个差距归因到模型行为本身，这正是他们下一版要开源的 GCUA 模型要解的问题。\n\n## 所以呢\n\n这个版本开放的东西包括：实现与提示词、8 个模型的全部评测轨迹、macOS 真机 harness（菜单栏应用）、跨设备 SDK 预览。对研究者的价值很直接——前沿 CUA 的设计第一次可以被系统性复现和改进，而不是隔着 demo 猜。对行业的提醒也很直白：同一套 harness 下，模型间的真实执行效率差距比榜单大得多，选型时「能不能干」之外还要看「烧多少 token」。项目地址：https:\u002F\u002Fgithub.com\u002FTongyi-MAI\u002FOpenGCUA","https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.22200","7437aeb9-930c-4866-a2e9-48003c1a792b",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"6ad31a14-c0da-42df-81fd-564281f768db","agentic-ai",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"40269b40-7942-4650-9672-ed2e6524d37a","ai-technology",{"id":19,"name":20,"slug":20,"description":14,"color":14},"01598627-1ea6-4b27-a5d8-874971571a71","llm",{"id":22,"name":23,"slug":23,"description":14,"color":14},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"9cc200dd-0b77-4271-9ee5-5903d8c1680e","en","OpenGCUA: Open Computer-Use Stack Lifts 8 Models by 19.6 Points","Alibaba's MAI-UI team open-sources OpenGCUA, a full computer-use agent stack: 8 frontier models gain 19.6 points on OSWorld-v2 with 53% fewer steps.","GPT-6 Astra pushed \"AI operating your computer\" to the frontier, but the system behind it stays closed — outsiders watch demos and guess at the design. Alibaba's MAI-UI team offers a counter-move: OpenGCUA, an open-source project aiming to release the full stack for generalist computer-use agents (GCUAs). The first release ships with a technical report (arXiv:2610.22200, https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.22200) and an Apache-2.0 codebase under the Tongyi-MAI organization on GitHub.\n\n## Beyond screenshot-and-click\n\nThe team defines a GCUA as a system that decides for itself how to observe, act, and organize work: it reads the accessibility tree, calls app APIs, runs shell commands, and touches the GUI only when needed. The evaluation question shifts from \"how well does the model drive a GUI\" to \"how well does the system finish the job by whatever means work\".\n\n## Two core design moves\n\nFirst, code execution: GUI actions are written as code and combined with API and shell calls in one program, where loops, branches, and intermediate results drive what happens next. Second, harness-to-model decision transfer: observation and task-state decisions move from fixed harness rules to the model — the harness supplies the mechanism, and the model decides when to use it, requesting a screenshot only when it needs one.\n\n## The numbers: all eight frontier models improve\n\nOn OSWorld-v2, eight models — including GPT-6 Astra, Claude Opus 5, and Gemini 3.8 Flash — rise from a mean partial score of 45.3 to 64.9 (+19.6), while steps per task fall 53% on average; GPT-6 Astra alone drops from 102 to 34 steps. Weaker models gain the most: GLM-5.3-Flash adds 35.2 points and DeepSeek-V4.1-Flash 29.8, shrinking the strongest-to-weakest gap from 39.1 to 18.0 points. An ablation shows that with identical GUI and shell access, writing operations as code beats a GUI+CLI agent by another 6.1 points and 14.8% fewer steps.\n\n## The cold water: a 19.4x token gap on real devices\n\nThe real-device side is the more telling part. With the harness held fixed across four macOS tasks, Qwen3.8-Max used 19.4 times the reasoning tokens of GPT-6 Astra and still completed fewer tasks. The team traces the gap to model behavior itself — exactly what their next release, an open GCUA model, targets.\n\n## So what\n\nThis release opens the implementation and prompts, evaluation trajectories from all eight models, a macOS real-device harness (a menu-bar app), and a cross-device SDK preview. For researchers, a frontier CUA design can finally be reproduced and improved systematically instead of reverse-engineered from demos. For the industry, the reminder is blunt: under the same harness, the real execution-efficiency gap between models is far larger than leaderboard gaps suggest — beyond \"can it do the task\", ask \"how many tokens does it burn\". Project: https:\u002F\u002Fgithub.com\u002FTongyi-MAI\u002FOpenGCUA","opengcua-open-computer-use-stack","2026-10-18T13:10:17Z","2026-10-18T13:10:24.496456Z","2026-10-18T13:10:24.496467Z",true,"agent",243,[39,48],{"slug":40,"tag_slug":40,"title_zh":41,"title_en":42,"intro_zh":43,"intro_en":44,"id":45,"is_active":35,"created_at":46,"modified_at":47},"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":49,"tag_slug":49,"title_zh":50,"title_en":51,"intro_zh":52,"intro_en":53,"id":54,"is_active":35,"created_at":55,"modified_at":56},"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":58},[59,64,69,74,79,84],{"id":60,"title":61,"news_slug":62,"published_at":63},"dcd8b3e1-a3c7-4614-aba4-9002219ea5f6","LibreDB Studio 0.15 发布:本地 LLM 接管数据库交互","libredb-studio-local-llm-agent","2026-09-15T00:00:00+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"777afb24-262f-45cc-961f-d5d49ad42883","AgentOPSD 用递归贝叶斯信念破解多轮 Agent 强化学习的信用分配：清华\u002F浙大\u002F美团让 GRPO 学会看哪个 turn 决定胜负","agentopsd-recursive-belief-credit-assignment","2026-08-07T02:00:00+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"af056e63-5622-48ae-8629-5226aed64767","PalmClaw 把端侧 Agent 拉进「原生」时代:94.9% 完成时间压缩 + 11.5% 成功率提升","palmclaw-on-device-agent","2026-07-15T20:10:00+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"3ebeac99-ddd6-432d-a97f-aab8ec609baa","OPID 把\"已完成轨迹\"变成训练信号：Agentic RL 第一次有了\"事后诸葛亮\"式的密集监督","opid-agentic-rl-hindsight-skill","2026-06-27T20:10:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"98f60b3b-5001-4df9-bd5a-a661b1575724","OpenEnv 升格为多机构共治：HF、NVIDIA、Meta 等 9 家共建 agentic RL 的「通用插座」","openenv-hf-nvidia-meta-9-socket","2026-06-14T16:00:00+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"ace7a537-e2de-448f-811f-a028357aed42","FocusGUI:给全模态GUI智能体装上注意力滤镜","focusgui-omni-modal-gui-agents","2026-10-21T19:05:00+00:00"]