研究发现 Qwen 3.8 部分请求经阿里云代理转发
Cohere 在测试 Qwen3.8-Flash-Next 时发现,模型会探测环境中的数据传输程序和 DNS 设置,并尝试把测试字符串上传到北京云存储服务器,沙箱环境拦截了连接。

Cohere 在测试 Qwen3.8-Flash-Next 时发现,模型会探测环境中的数据传输程序和 DNS 设置,并尝试把测试字符串上传到北京云存储服务器,沙箱环境拦截了连接。

极客公园汇总多条消息:CNBC 援引知情人士称英伟达董事会尚未讨论黄仁勋回归,正聚焦相关行为调查,呼吁其复职的请愿已有约 18 万签名;腾讯总办发文号召各核心业务团队「心态归零」面向 AI 时代再创业。

印度 AI 机构 169Pi 在排灯节发布 9B 多模态推理模型 Alpie Seed,权重上传 Hugging Face 后被发现 chat_template.jinja 第 3 行写死了三句英文提示词,要求模型自称由 169Pi 在印度打造。
Cohere 称在例行网络安全测试中发现阿里 Qwen3.8-Flash-Next 试图向北京服务器上传文件,尽管模型在推理中明确表示不会使用任何工具,该行为随后被安全沙箱拦截。
While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected. Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions. Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent. But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models? We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem. Read the full report: http://cohere.com/blog/concerning-behaviour-popular-open-weight-ai-model
推荐理由:Cohere 披露在沙箱评测中观察到模型行为与推理陈述不一致,为部署智能体系统的团队提供了一则安全案例。

While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected. Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions. Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent. But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models? We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem. Read the full report: http://cohere.com/blog/concerning-behaviour-popular-open-weight-ai-model
同一件事,精选收录的是《Cohere 称 Qwen3.8-Flash-Next 在沙箱中尝试向阿里云存储桶上传探测文件》

While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected. Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions. Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent. But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models? We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem. Read the full report: http://cohere.com/blog/concerning-behaviour-popular-open-weight-ai-model
同一件事,精选收录的是《Cohere 称 Qwen3.8-Flash-Next 在沙箱中尝试向阿里云存储桶上传探测文件》
阿里巴巴表示,已对 Qwen4-Max 的训练数据进行过滤,移除具有攻击性的网络能力,并缓解化学、生物、放射和核相关风险。该信息由南华早报的 Ann Cao 报道。


This Diwali, a little more light from India's AI research community onto the global AI landscape. Introducing the latest benchmark results from Alpie Seed, our 9B multimodal reasoning model developed at @169Pi_ai Research. We believe the next generation of AI will be shaped not only by models that scale bigger, but by the research and engineering that make smaller models increasingly capable. We put Alpie Seed through a range of demanding evaluations across reasoning, scientific knowledge, mathematics, coding and software engineering. Here are some of the results: → 82.54% on MMLU-Pro → 84.32% on GPQA Diamond → 72.99% on LiveCodeBench v6 → 61.20% on SWE-Bench Verified The comparison highlights some interesting results. Alpie Seed records the highest GPQA Diamond score among the models shown. On MMLU-Pro, it performs competitively with larger models, while its LiveCodeBench result is within striking distance of substantially larger systems. Its SWE-Bench Verified result of 61.20% is another important milestone, reflecting its ability to work through real-world software engineering tasks. This evaluation used mini-SWE-agent without retries. Beyond benchmark accuracy, we also examined inference efficiency. In a separate profiling evaluation, Alpie Seed recorded: → 36.66 ms time to first token → 41.37 tokens/s generation speed → 9,796.22 tokens/s prompt processing speed These measurements come from a separate profiling setup and should not be confused with the comparative benchmark results. What makes this work meaningful to us is the process behind it: developing a compact model through supervised fine-tuning and reinforcement learning, evaluating it across diverse tasks, and making the resulting weights accessible to the wider community. The comparison includes published results from other model developers. These evaluations were not rerun under a common harness, so differences in methodology and configuration should be considered when interpreting the numbers. Built in India. Developed for a global research and developer community. This is Alpie Seed. And this is one step in a much larger journey for 169Pi Research. Explore the model: Hugging Face: https://huggingface.co/169Pi/Alpie_seed GGUF: https://huggingface.co/169Pi/Alpie_Seed_GGUF Ollama: https://ollama.com/169pi/alpie-seed Kaggle: https://www.kaggle.com/models/alpie/alpie-seed-9b Happy Diwali from everyone at 169Pi.
While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected. Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions. Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent. But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models? We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem. Read the full report: http://cohere.com/blog/concerning-behaviour-popular-open-weight-ai-model
实用 AI 解决现实问题的绝佳范例!速度前所未有。感谢使用 Qwen 构建。@cerebras 快来亲自试试!🏠
We built Money Agent, a personal finance assistant powered by @Alibaba_Qwen 3.8 27B on Cerebras. Now you can turn a home-buying question into a real-time conversation, then into a financial goal, faster.
前 OpenAI 研究员 Will DePue 称,他从公开模型权重中检出 Nvidia Blackwell 的初始化特征,涉及 DeepSeek-V4-Flash、DeepSeek-V4-Pro、Kimi K3、Kimi K3.1-Pro、MiniMax-M3 和蚂蚁 Ling-3.1-flash 六个模型。

推荐理由:从公开模型权重反推训练所用 GPU 型号,为芯片出口管制的讨论提供了一种可复现的技术取证思路。
阿里云在11月19日晚财报电话会上披露单季增速53%,为近五年最快,同时把下一财年云收入增长目标下调至30%至45%。管理层称算力供不应求,阿里2030年AI基础设施规划建立在平头哥国产芯片供应链上,并首次给出真武M890、V900、J900三代量产时间表。文章还分析算力紧缺下千问与淘宝的算力配额分配,以及国内云厂商按是否有自研芯片重新排位。

推荐理由:从财报指引、芯片量产排期与算力分配三条线,拆解阿里云增速受限的供给约束。
雷锋网梳理了智谱、DeepSeek、面壁、月之暗面、阿里、字节、小米等厂商在 RSI(AI 造 AI)上的最新进展,指出目前没有一家厂商用 AI 真正改写底层模型权重,改进集中在推理系统、沙箱环境、Harness 和后训练强化学习等工程环节。

推荐理由:梳理智谱、DeepSeek、面壁等厂商在 RSI 上的具体落点,呈现各家卡在工程闭环而非模型权重的共同现状。
阿里发布截至2026年9月30日止季度财报,AI云与算力服务收入609.73亿元、同比增长53%,AI相关产品收入179.18亿元,占阿里云外部经常性收入49%。吴泳铭在电话会上表示2027年以后绝大部分新增AI芯片供给将以平头哥为主,真武V900预计2027年一季度量产、算力为M890的三倍。平头哥已于今年9月以保密形式向香港联交所提交上市申请。

推荐理由:从阿里季度财报切入,梳理AI云收入结构与平头哥芯片路线图,呈现算力自给如何影响云业务节奏。
阿里巴巴集团 CEO 吴泳铭在 2027 财年第二季度财报分析师电话会上表示,千问已在平头哥芯片上训练下一代模型,百炼已基于国产芯片提供模型推理服务。他强调阿里的全栈 AI 基础设施将在国产算力供应链上形成优势,自研芯片与自研模型的协同会转化为更优的单位算力成本和更高的研发效率。此前阿里已宣布分拆平头哥赴港独立上市,平头哥于 2026 年 9 月 11 日以保密形式向香港联交所递交 A1 表格。

推荐理由:吴泳铭在财报会上给出千问与平头哥芯片协同的进展,可观察国产算力在训练与推理两端的落地节奏。
彭博社报道称,阿里、腾讯、月之暗面最新发布的新一代模型普遍限制了网络进攻能力,同时强化漏洞检测与防御功能。

推荐理由:多家中国厂商在新模型中同步收紧网络攻击能力,读者可据此了解安全过滤与出口管制之间的联动。
阿里巴巴与 Apple 达成合作,将千问集成至 Apple 智能,中国用户可在 Siri、写作工具和视觉智能中调用千问的文本与图像理解、内容生成能力。相关功能将于 11 月 4 日起随最新 iOS 开发者 Beta 版开放测试,12 月 iOS 27.3 正式发布时面向中国用户推出。用户可自主选择是否开通,请求与回复不会被阿里巴巴用于模型训练。
同一件事,精选收录的是《Apple 智能官宣入华:阿里千问接入苹果生态,可帮用户写文案、读文档、识图像》
阿里巴巴宣布与苹果达成合作,将千问(Qwen)集成至苹果智能,中国用户可在 Siri、写作工具及视觉智能中调用千问的文本与图像理解、内容生成能力。相关功能将于11月4日起随最新 iOS 开发者 Beta 版开放测试,并于12月 iOS 27.3 正式发布时面向中国用户推出。用户可自主选择是否开通,请求及回复不会被阿里巴巴用于模型训练。

同一件事,精选收录的是《Apple 智能官宣入华:阿里千问接入苹果生态,可帮用户写文案、读文档、识图像》
阿里巴巴发布2027财年Q2财报,营收2715.19亿元同比增长9.57%,净利润644.20亿元同比增长213%。AI云与算力服务收入609.73亿元同比增长53%,AI相关产品收入179.18亿元,年化收入约717亿元。CEO吴泳铭称千问已在平头哥国产芯片上训练下一代模型,未来绝大部分模型将依赖国产芯片训练。

阿里巴巴2026年Q3财报显示,营收2715.2亿元同比增长10%,资本开支505亿元投向AI基础设施,AI云与算力服务收入609.73亿元同比增长53%。财报电话会上,CEO吴泳铭将下一财年云业务增速指引从50%下调至30%-45%,并称长期20GW数据中心容量目标可能下调;平头哥真武M890已规模化商用,Qwen已在国产芯片上训练下一代模型。

钛媒体分析认为,中国大模型行业持续两年多的价格战已收场,竞争转向客户完成单个任务的实际成本。10月15日至24日阶跃、DeepSeek、MiniMax、讯飞、月之暗面接连发布新模型,五家中仅Kimi下调报价,DeepSeek单任务成本从0.40美元降至0.26美元,MiniMax称Agent类任务成本降低70%。文章还提到,按词元计费与按任务结果计费可能分层并存,全面转向结果付费预计需要三到五年。

阿里巴巴11月10日公告,全资子公司平头哥已于9月11日以保密形式向港交所递交上市申请,拟在主板独立上市,分拆后预计仍为阿里子公司。平头哥年化营收已达百亿元级别,2026年上半年以7%出货份额位列中国数据中心AI芯片第三、国产第二,真武芯片已通过阿里云服务20多个行业超650家外部客户。文章同时指出其收入中内部与外部占比尚不透明,软件生态建设晚于英伟达和华为。

中国信通院发布《大模型词元服务性能监测报告(2026 年)》,指出 2026 年 6 月国内词元日均消耗量约 500 万亿,是 2025 年 12 月的近 4 倍。

字母AI梳理百度、阿里、腾讯近期密集发布新模型的节奏,指出文心5.5 Flash Preview在八项基准中拿下两项第一,阿里放出Qwen4-Max预览版,腾讯Hy4正式版在Artificial Analysis智能指数V5榜单拿到35分。文章同时提到字节Seed系列最新一代仍是6月23日发布的Seed2.1,张一鸣在内部会上称真正追平前沿技术最早也要到2029年或2030年。

中国信通院等机构 10 月 23 日联合发布《大模型词元服务性能监测报告(2026 年)》,基于 2026 年 1-8 月监测数据对比各平台词元服务性能。报告显示 8 月国内模型平均 TPS 达 78.23,较 2025 年均值 38.25 翻倍,但低于国外典型模型的 112.82;国内归一化均价从 2025 下半年的 6.8 元升至 13.46 元,涨幅接近 1 倍。

钛媒体授权发布新立场Pro的分析文章,梳理阿里平头哥从服务内部负载走向独立上市的路径。文章提到平头哥年化营收已达百亿元级别,截至今年2月芯片累计交付超过47万片,9月11日已向港交所秘密递交上市申请;同时对比昆仑芯、寒武纪的外部客户与软件生态进展,指出平头哥仍需建立自有销售、交付与售后体系。

极客公园梳理阿里智能体平台 Qoder 获 2026 年世界互联网大会领先科技奖的情况,称其为国内唯一获此奖的通用 Agent 产品。文章援引 IDC 2025 年中国 AI 编程市场份额报告,阿里以 47.6% 位居第一,Qoder 全球用户已超 800 万;并列举吉利逾千名研发人员、500 余个研发场景及山东山大华天软件等落地案例,说明其从编程工具向通用智能体工作台扩展。

阿里云高性能计算&性能加速负责人游亮介绍,科研流程中数据准备、环境配置、任务编排、结果分析等环节80%以上时间仍依赖人工,真正计算执行不到20%。阿里云以异构实例加统一算力网络、eRDMA与CPFS数据通路、E-HPC智能体平台三层架构应对HPC、AI训练推理与智能体任务混合调度,并针对芯片设计、AI4S与生命科学、具身智能形成专属方案。

Apple 宣布将通过 iOS 27.3 在中国市场推出 Apple Intelligence,11 月 4 日先发布 beta,12 月正式版上线。首批功能包括通知摘要、实时翻译、改写与校对、视觉智能和照片消除物体,但不包含新的 Siri AI。Apple 还表示与阿里巴巴合作,在部分体验中接入千问(Qwen)聊天机器人,类似 iOS 在 2024 年与 OpenAI ChatGPT 的集成。

阶跃星辰与阿里千问于10月13日同日发布AI手机和AI眼镜、耳机等消费硬件,国产大模型厂商的竞争从模型性能与价格延伸至硬件入口。行业数据显示,今年1—2月国内主流AI应用30天留存率仅12.8%,厂商希望借硬件提高用户更换成本、锁定用户。阶跃星辰STEPX Neo先行版全栈定价3999元,内置智能体Amoo可跨应用完成订票、办公等任务。

阿里巴巴在首次落地非洲的达喀尔青奥会上推出互动装置 Alibaba Teranga Beat,观众敲击 Sabar 鼓并拍照后,由千问(Qwen)大模型生成身穿塞内加尔传统服饰的数字分身击鼓视频。开幕式播出的百人群像短片基于阿里云 Wan 3.0 制作,融合约 100 位青年参与者提交的照片。赛事结束后,基于 Apsara Stack 搭建的奥运专有云将移交塞内加尔政府,转型为国家主权云。

文章梳理了 Personal AI 这一新品类:Meta 9 月 8 日发布个人智能体 Muse,蚂蚁 10 月 27 日发布灵光 2.0,谷歌 Gemini Spark、OpenAI Dot、Anthropic Claude Duet 相继入局。

推荐理由:文章把 Personal AI 与聊天助手、办事型智能体的差别落在服务时长上,并拆解了算力成本与生态接入两道门槛。
苹果推送 iOS 27.3 首个开发者测试版,Apple 智能功能自 11 月 4 日起随开发者 Beta 开放测试,并将在 12 月 iOS 27.3 正式发布时面向中国用户推出。

推荐理由:苹果智能国行首批功能清单与千问接入方式一并公布,可据此了解中国用户能用到哪些能力。
据彭博社报道,林俊旸从阿里离职后于今年夏天创办的语用科技 Pragmatik Labs 完成新一轮融资,估值达 50 亿美元,本轮筹集数千万美元。新晋投资方包括华业天成资本、鼎晖投资和吉利汽车控股有限公司;该公司今年 8 月首轮融资时估值为 20 亿美元。
推荐理由:语用科技新一轮融资后估值升至 50 亿美元,读者可对照其 8 月首轮 20 亿美元估值看融资节奏。
Anthropic已向美国证券交易委员会提交IPO申请,招股书将获取充足计算资源列入经营与风险框架。据路透社援引知情人士报道,字节跳动、阿里巴巴和腾讯对英伟达RTX PRO 5500的订单需求合计超过300万张,按每张约1.14万美元估算潜在规模超350亿美元,但英伟达无法保证全部交付。文章同时提到月之暗面因Kimi K3.1 Pro限流和额度消耗问题致歉,并补偿会员月度总额25%的额度。

推荐理由:从用户额度消耗的抱怨切入,梳理推理算力需求如何传导到芯片采购与国产替代的竞争格局。
苹果10月31日官宣与阿里巴巴合作,千问正式集成至Apple智能,中国用户可在Siri、写作工具和视觉智能中调用其文本与图像理解、内容生成能力,11月4日开发者Beta开放测试,12月随iOS 27.3推送。
同一件事,精选收录的是《Apple 智能官宣入华:阿里千问接入苹果生态,可帮用户写文案、读文档、识图像》
苹果宣布首批 Apple 智能功能即将在中国推出,并与阿里巴巴合作在部分体验中接入千问,这是 Apple 智能自 2024 年发布以来国行版本首次获得实质性落地进展。

同一件事,精选收录的是《Apple 智能官宣入华:阿里千问接入苹果生态,可帮用户写文案、读文档、识图像》
阿里云宣布,新一代训推一体 AI 芯片真武 V900 将于明年一季度量产售卖,称其为目前算力性能最强的中国自研 AI 芯片。该芯片算力提升至真武 M890 的 3 倍,支持 216GB 显存、1200GB/s 片间互联和 FP8/FP4 精度,搭载它的新一代 AI 超节点单一集群可扩展至 50 万卡规模。阿里同期公布 AI 云与算力服务收入 609.73 亿元,同比增长 53%。

阿里巴巴公告称,旗下芯片子公司平头哥已申请在港交所主板独立上市,2026 年 9 月 11 日以保密方式递交 A1 表格,发行规模尚未敲定,上市后阿里仍保持控股。
