[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-cua-s1-forms-system-one-model":3,"topics-all":38,"news-related-0565190a-0bcd-492f-934f-0ad2ab32f485":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},"0565190a-0bcd-492f-934f-0ad2ab32f485","70万参数2.8MB填一张表:Cua开源CUA-S1,单次前向替代23轮LLM","Cua 团队开源 CUA-S1 家族首个模型 CUA-S1-FORMS:706k 参数、2.8MB、单次前向 50 毫秒完成整表决策,官方自报真实表单准确率 99.7%,对照通用 LLM API 基线约 83.6%。MIT 许可,CPU 即可运行。","填一张网页表单,今天的通用 LLM Agent 要走多少步?Cua 团队给的参照数字是 23 轮、39.6 秒——截图、思考、调用驱动器、再截图。而他们刚开源的专用模型 CUA-S1-FORMS 做同一件事,只需要一次 50 毫秒的前向传播。\n\n## 70 万参数的「系统一」模型\n\nCua(YC S25 团队)9 月 19 日开源了 CUA-S1 家族的首个模型 CUA-S1-FORMS。家族命名叫 System One Models:不做链式思考、不做多轮推理、不调云端 API,单个前向传播直接给出决策。CUA-S1-FORMS 只有 706k 参数,checkpoint 2.8MB,MIT 许可,权重发布在 Hugging Face 的 cua-ai\u002Fcua-s1-forms,CPU 上就能跑。\n\n它只干一件事:表单决策。给定一组从文档里抽出的结构化字段和候选值,模型对每个表单元素预测四个动作之一——填入给定值、勾选、点击、跳过。所有元素的决策在一次前向里并行完成,执行顺序由调用方代码排好,交给 Cua Driver 逐个静默执行。它不预测新值,也不看截图。\n\n## 架构:小到反直觉\n\n第三方拆解给出的架构细节:两层、128 宽度的 transformer,4 个注意力头,直接读原始字节、不用 tokenizer。训练成本也轻得不像一次模型发布:第一版在合成数据上训练不到 30 分钟。\n\n## 自报成绩与边界\n\n官方 model card 自报:真实表单决策准确率 99.7%,作为对照的通用 LLM API 基线约 83.6%。注意这是官方自报数字,独立复现还没出现。能力边界同样明确:只在任务边界清晰、结构受限的场景成立,预测新值、理解截图这类开放任务它明确不做——这些仍留给大模型。\n\n## 行业读法:小模型开始「分食」Agent 流水线\n\n同期 Latent.Space 的 AI 周报提到,@ValsAI 发布的 CUA-Bench 上所有前沿模型在实时键鼠操控任务里得分都低于 20%——通用 Agent 在计算机操控这件事上远未饱和。这个背景下,CUA-S1 走的是一条反向路线:不把宝押在更大的通用模型上,而是把 Agent 流水线里「反射级」的子任务切出来,交给训练不到半小时的微型模型。前一天开源的 Needle 3(8-29MB 端侧工具调用模型)是同一叙事的另一个数据点。\n\n对做企业自动化的团队,信号很具体:不是 Agent 的每个环节都值得挂一个千亿模型。把「值对齐」这类确定性子任务下沉到 2.8MB 的本地模型,延迟、成本、隐私三笔账同时改善。至于官方数字的成色,等独立评测说话。","https:\u002F\u002Fgithub.com\u002Ftrycua\u002Fcua","998df6db-96e6-4b8e-8be1-cfa00a6cd177",[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},"7ac06d8e-b074-4147-abfc-ffaa4c6b8744","ai-efficiency",{"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},"d6d187b1-6a42-4a77-942d-66a88e6acda0","en","CUA-S1: Cua's 706K-Param Model Fills a Form in One 50ms Pass","Cua open-sources CUA-S1-FORMS: a 706K-param, 2.8MB model that fills a form in one 50ms pass, replacing a 23-turn LLM loop. Self-reported 99.7% accuracy.","How many steps does a general-purpose LLM agent take to fill one web form? Cua's reference number: 23 turns and 39.6 seconds — screenshot, think, call the driver, screenshot again. Their newly open-sourced specialized model, CUA-S1-FORMS, does the same job in a single 50-millisecond forward pass.\n\n## A 706K-Parameter \"System One\" Model\n\nCua (a YC S25 team) open-sourced CUA-S1-FORMS on September 19, the first release in its CUA-S1 family. The family is named System One Models: no chain-of-thought, no multi-turn reasoning, no cloud API call — one forward pass produces the decision. CUA-S1-FORMS has just 706k parameters, a 2.8MB checkpoint, an MIT license, weights published on Hugging Face as cua-ai\u002Fcua-s1-forms, and it runs comfortably on CPU.\n\nIt does exactly one job: form decisions. Given a set of structured fields and candidate values extracted from a document, the model predicts one of four actions for each form element — use the given value, check, click, or skip. All element decisions come back in parallel from a single forward pass; your code orders the actions, and Cua Driver executes them silently one at a time. It does not predict new values, and it does not look at screenshots.\n\n## Architecture: Counter-Intuitively Small\n\nThird-party teardowns describe a two-layer, 128-width transformer with four attention heads that reads raw bytes instead of using a tokenizer. Training was equally light: the first iteration took less than 30 minutes on synthetic data.\n\n## Self-Reported Results and Limits\n\nThe official model card reports 99.7% accuracy on real form-filling decisions, against roughly 83.6% for a general-purpose LLM API baseline. Note these are vendor-reported numbers; independent replication has not appeared yet. The limits are equally explicit: this works only where task boundaries are clear and structure is constrained. Predicting new values or understanding screenshots — the open-ended work — stays with large models.\n\n## Industry Read: Small Models Start Carving Up the Agent Pipeline\n\nThe same week, Latent.Space's AI newsletter noted that on @ValsAI's CUA-Bench, every frontier model scores below 20% on real-time keyboard-and-mouse tasks — general agents are far from saturating computer control. Against that backdrop, CUA-S1 takes the reverse route: instead of betting on ever-larger general models, it carves the \"reflex-level\" subtasks out of the agent pipeline and hands them to a micro-model trained in under half an hour. Needle 3, the 8-29MB on-device tool-calling model open-sourced a day earlier, is another data point in the same narrative.\n\nFor teams building enterprise automation, the signal is concrete: not every step of an agent deserves a hundred-billion-parameter model. Sinking deterministic \"value alignment\" subtasks into a 2.8MB local model improves latency, cost, and privacy at the same time. As for the vendor numbers — wait for independent evaluation.","cua-s1-forms-system-one-model","2026-09-20T13:11:48Z","2026-09-20T13:11:56.294327Z","2026-09-20T13:11:56.294340Z",true,"agent",8,[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},"13378d5e-2440-496d-8c3c-7d36858e641d","不聊天的端侧基座:Needle 3 用 8-29MB 在微控制器上跑工具调用","needle-3-tiny-tool-calling-model","2026-09-19T13:09:46+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"259d91b2-ed6b-4af8-8f2a-f759b84cc617","蚂蚁 Ling-3.0 Flash：124B\u002F5.1B MoE 的 Agent 生产级模型","inclusionai-ling-3-flash-hybrid-linear-moe-agent","2026-08-14T08:00:00+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"9ba1770e-87f2-47b9-aaa1-19f4f2ba78f1","ScienceIDE:把全球科学代码变成智能体训练场","scienceide-scientific-code-agent-environments","2026-09-17T23:05:17+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"4c4a2a9e-f69b-4985-bd42-97ab2ef4e2ac","Spark-X2.5-4B 开源:4B 跑 1M 上下文,22 项基准打 9B 级 Qwen3.5","spark-x2-5-4b-apache-open-source","2026-09-16T01:30:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"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":85,"title":86,"news_slug":87,"published_at":88},"dcbb1042-4fef-4a6a-bd7a-0e11b4d97c10","TRL v1.13.0:一处 dtype 修复拿回 21.6% kernel 时间,PPOTrainer 退役","trl-v1-13-tensor-core-lmhead-ppotrainer","2026-09-12T21:05:00+00:00"]