[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-shanghai-ai-lab-atria-dawn-preview":3,"topics-all":38,"news-related-28e1fd65-7f26-41e8-b1ba-aa69d03dce84":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},"28e1fd65-7f26-41e8-b1ba-aa69d03dce84","上海AI实验室开源 Atria Dawn Preview:五项 Agent 基准居首","上海AI实验室开源 Atria Dawn Preview:744B GLM-5.2 MoE 基座,256K 上下文,MIT 协议,另有 FP8 版。官方 16 项对照 5 项居首,CyberGym、BrowseComp 等 agent 基准领先,SWE-bench Pro 等工程基准落后 Opus 5。","今天凌晨,上海人工智能实验室把新一代 agentic 模型 Atria Dawn Preview 的权重放上了 Hugging Face 和 ModelScope——MIT 协议,基于 744B 参数的 MoE 基座 GLM-5.2,上下文 256K,同步放出 FP8 量化版。它不做聊天人设,主打一件事:把开放式科研任务推进到\"可执行、可验证、可复现\"的结果。\n\n## 把\"验证\"做进训练循环\n\nAtria 的核心是一条 Verifiable Experience Pipeline:每个训练任务都接真实执行环境,模型观察状态、调用工具、产出中间产物、根据反馈修正;最终结果靠外部信号验证——测试通过、指标、文件状态、几何结构、来源证据。只有同时连着\"任务-轨迹-产物-验证证据\"的经验,才被吸收为模型的可复用能力。官方把它总结成三分工:模型内化过往经验,harness 管理当下流程,环境决定每个动作的后果。这相当于把\"agent 能力靠 prompt 工程外挂\"的老路,换成训练期内化。\n\n## 跑分:Agent 基准多项居首,工程交付仍是短板\n\n官方对照表拉了六个对手:DeepSeek V4 Pro 0813、Kimi K3、Qwen 3.8 Max、GLM 5.3、GPT 5.6 sol、Claude Opus 5。16 项基准里,Atria 在 5 项拿到行内最高:AutomationBench 53.8(次高 49.7)、BFCL v4 77.0、CyberGym 86.5、DeepSearchQA 96.0、BrowseComp 92.5。浏览器式深度检索和网络安全这两类长程 agent 任务,优势最明显。\n\n但同一张表也自曝短板:SWE-bench Pro 59.6,Opus 5 是 74.7;JobBench 50.3 对 68.0;Terminal-Bench 2.1 78.3 对 90.2。科研探索型任务强、工程交付型任务弱,格局非常清楚——官方没有回避,这比通稿式发布可信得多。\n\n## 一个 demo:从 100GB 气象数据到全球预报系统\n\n官方展示里最硬的一条:禁用联网的环境下,给模型 100GB 以上的全球天气数据,它自己设计出 0.4B 以上参数的 ViT 网络,训练 45,000 步,建模 69 个气象变量的演化,最终系统一分钟内完成未来一周的全球预报,部分预测指标超过 NVIDIA 的经典预报模型 FourCastNet(官方自报)。另有 20 分钟从零搭出 MiniOS 的演示。\n\n## 所以呢\n\nPreview 阶段、纯文本输入、自报对照表,这些折扣都该打。但两点值得盯:第一,744B MoE 级别的权重 MIT 全量开源,部署文档直接给了 SGLang、vLLM、Codex、Claude Code 四条接入路径;第二,当 agent 竞赛从聊天跑分转向任务交付跑分,验证信号从哪来决定含金量——Atria 把验证做进训练循环,这个方向比榜单数字本身更重要。\n\n参考:[Atria 官网](https:\u002F\u002Fatria-asi.ai\u002F) · [HF model card](https:\u002F\u002Fhuggingface.co\u002Finternlm\u002FAtria-Dawn-Preview)","https:\u002F\u002Fatria-asi.ai\u002F","39280ede-8490-453c-a90d-44caf44eb6bc",[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},"7e89b5cc-57db-4f37-bc6d-28919a73931c","model-release",{"id":19,"name":20,"slug":20,"description":14,"color":14},"d11f0044-8aef-487c-bebe-89ce4683a4a3","moe",{"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},"06899a8e-d80a-4f8a-8b84-de5344177d61","en","Shanghai AI Lab releases Atria Dawn Preview, a 744B MoE agent","Shanghai AI Lab open-sources Atria Dawn Preview: 744B MoE agent, MIT weights, FP8. Tops 5 of 16 official benchmarks, trails Opus 5 on SWE tasks.","Shanghai AI Laboratory has posted the weights of Atria Dawn Preview, its new-generation agentic model, on Hugging Face and ModelScope — MIT license, built on a 744B-parameter MoE GLM-5.2 foundation, 256K context, with an FP8-quantized variant released alongside. The model skips the chatbot persona entirely; its one pitch is pushing open-ended research tasks toward results that are executable, verifiable, and reproducible.\n\n## Verification built into the training loop\n\nAt Atria's core is a Verifiable Experience Pipeline: every training task connects to a real execution environment, where the model observes state, calls tools, produces intermediate artifacts, and adapts to feedback. Final outcomes are verified through external signals — passing tests, metrics, file state, geometric structure, or source evidence. Only experience that ties together the task, its trajectory, its artifacts, and verification evidence gets absorbed as reusable capability. The team summarizes this as a three-way split: the model internalizes past experience, the harness organizes the process at hand, and the environment determines the consequences of each action. In effect, it replaces the old \"bolt agent skills on with prompt engineering\" route with internalization during training.\n\n## Benchmarks: agent tasks lead, engineering delivery lags\n\nThe official comparison table pulls in six rivals: DeepSeek V4 Pro 0813, Kimi K3, Qwen 3.8 Max, GLM 5.3, GPT 5.6 sol, and Claude Opus 5. Across 16 benchmarks, Atria posts the highest score in its row on five: AutomationBench 53.8 (next best 49.7), BFCL v4 77.0, CyberGym 86.5, DeepSearchQA 96.0, and BrowseComp 92.5. Long-horizon agent work — browser-style deep search and cybersecurity — is where the lead is widest.\n\nThe same table also exposes the weaknesses: SWE-bench Pro 59.6 against Opus 5's 74.7; JobBench 50.3 versus 68.0; Terminal-Bench 2.1 78.3 versus 90.2. Strong on research-exploration tasks, weak on engineering-delivery tasks — a clear profile that the official page does not hide, which counts for more credibility than a press-release launch.\n\n## One demo: from 100GB of weather data to a global forecast system\n\nThe hardest demo on the site: in an environment with web search disabled, given more than 100GB of global weather data, the model designed a ViT network with over 0.4 billion parameters, trained it for 45,000 steps, modeled the evolution of 69 weather variables, and produced a system that forecasts a week of global weather within a minute — outperforming NVIDIA's classic FourCastNet on some prediction metrics (per the official claim). A separate demo builds a MiniOS from scratch in 20 minutes.\n\n## So what\n\nIt's a preview, text-input only, and the comparison table is self-reported — take the discounts. Two things are worth watching. First, 744B MoE-class weights fully open under MIT, with deployment docs covering SGLang, vLLM, Codex, and Claude Code out of the box. Second, as the agent race shifts from chat scores to task-delivery scores, where the verification signal comes from decides how much a score is worth — Atria putting verification inside the training loop matters more than any single leaderboard number.\n\nRefs: [Atria site](https:\u002F\u002Fatria-asi.ai\u002F) · [HF model card](https:\u002F\u002Fhuggingface.co\u002Finternlm\u002FAtria-Dawn-Preview)","shanghai-ai-lab-atria-dawn-preview","2026-09-14T19:08:09Z","2026-09-14T19:08:14.480475Z","2026-09-14T19:08:14.480484Z",true,"agent",157,[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},"643addcf-6639-4c7b-b6b9-dc478bab4792","Nex-N2.5 三档齐发:1.6T MoE 权重开放,BrowseComp 92.6 超 Opus 5","nex-n2-5-open-weights-agent-family","2026-09-12T13:07:00+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"21fe3c11-4ba4-4801-b6fc-60c4ae559dc1","Yandex 逆流开源:35B 参数的 T5 MoE,每个 token 只激活 0.6B","yandex-aliceai-t5-sparse-moe","2026-09-16T19:11:43+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"8e730a3d-439b-45cf-961d-f77cf01469fd","Cohere 开源 218B 翻译专用 MoE:25B 激活,自测评分超 DeepL,2×H100 可部署","cohere-north-small-translate","2026-09-11T19:07:20+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"d941056b-c2e7-42e5-965a-a982c20b1169","Qwen3.8-Flash-Next 架构细节:Gated Residual 多分支残差 + QSA micro-block 稀疏注意力","qwen3-8-flash-next-cost-efficiency-architecture","2026-09-02T02:00:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"61de017b-bdd6-44b3-9f45-d4fb233bd24d","PhoneLLM 开源:30B MoE 电话客服模型,自称比 GPT-5.6 Terra 便宜 94%","phonellm-alpha-1-voice-agent-open-model","2026-08-29T21:10:00+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"33f3b08b-c8a2-43ec-81cf-85e2b918f913","腾讯开源 Hy4 preview:770B MoE、1M 上下文,模型首次参与自身训练","tencent-hy4-preview-770b-moe","2026-08-29T15:00:00+00:00"]