[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-four-ai-models-overlapping-outage-sept-2026":3,"topics-all":41,"news-related-1942b07b-f794-42b1-b944-ca6b32d4ae16":60},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":27,"news_slug":34,"published_at":35,"created_at":36,"modified_at":37,"is_published":38,"publish_type":39,"image_url":14,"view_count":40},"1942b07b-f794-42b1-b944-ca6b32d4ae16","四大 AI 模型同日集体掉线:OpenAI\u002FClaude 官方确认,Gemini\u002FGrok 表面沉默","9 月 3 日美东早高峰,ChatGPT、Claude、Grok、Gemini 几乎同时报告严重故障,Anthropic 与 OpenAI 公开承认并启动修复,Google 和 xAI 始终未发声明。AWS、Azure、Cloudflare 的状态页同样干净,但 DownDetector 数据印证了用户侧的波动。","## 一次罕见的\"四家同时掉线\"\n\n2026 年 9 月 3 日美东时间上午,ChatGPT、Claude、Grok、Gemini 这四款主流 AI 服务在不到三小时内接连出现严重服务中断。Ars Technica 援引各厂商状态页与 DownDetector 数据复盘了这一窗口,结论很直白:四款前沿模型在同一时段集中故障,这种事\"几乎前所未见\"。\n\n## Anthropic 先开口,9:23 起报告部分故障\n\n时间线上,Anthropic 是最早响应的厂商。根据 status.claude.com 的事件 461yvfrzpwtt,美东 9:23 报告 Claude Mythos 5.1、Claude Fable 5.1 与 Claude Opus 5 出现\"elevated errors\",约 15 分钟后定位原因,12:16 标注已修复,整个事件大约三小时。中午刚过,Claude Sonnet 5 又单独拉了一次事件(报告 288w7p4hk1l1),显示同一个工作日内多个模型家族先后承压。\n\n## OpenAI 10:43 跟进,12:55 收尾\n\nOpenAI 在 10:43 通过 status.openai.com 发布事件 01M1KWEDH417T2CF44YYHZDFCR,通报 ChatGPT 与 Codex 出现\"elevated errors\"和性能下降。半小时内缓解措施上线,12:55 标记为 resolved。Ars 提到 OpenAI 同期公布的 ChatGPT 90 天可用性为 99.63%、ChatGPT Codex 为 100%,意味着这次属于\"压在小数点后两位\"的罕见情况。\n\n## Grok 和 Gemini:沉默,但数据不会说谎\n\nxAI 的 Grok 在用户端直接弹出\"is experiencing issues\"错误提示,DownDetector 报告数从 9 点前的不足 10 条,在 9:45 冲到 1365 条,之后回落到 273。Google 方面,Gemini 从未公开承认故障,但 DownDetector 上 Gemini 报告数从 10:30 左右的 23 条在 11:00 跳到 412 条,StatusGator 也把 Gemini API 在 10:45 至 11:15 之间的状态标为\"likely outage\"。两家公司的策略一致:不在状态页留痕,把认领压力留给第三方监测平台。\n\n## 云厂商没事,但社区数据有波动\n\n事件中值得注意的是,AWS、Azure、Cloudflare 这三家最常被怀疑的底层云服务商,在自家状态页上没有公开重大故障记录。然而 DownDetector 显示这三家的报告数当日早间都出现过可见峰值,提示\"上层的服务故障可能短暂把流量挤到其他 SaaS、绕了一圈再回到云本身\"。Ars 把它解读为一种\"上层模型层出问题,下层公共云基本没出问题\"的分裂图景,这本身就是值得记住的信号。\n\n## 个人评论:这是\"模型层单点\"不是\"云层单点\"\n\n这一次事件最值得行业反思的是故障切片的位置。四个模型来自四个不同公司、跑在多家不同云上、用着不同的推理栈和 GPU 集群,但它们几乎同时出问题,意味着问题不太可能在物理基础设施层。更可能的解释有几条:一是大量企业级流量在背后接了多家模型的容灾切换,一家挂了流量被推到下一家形成连锁压力;二是某个上游供应商——比如 CDN、身份认证、DNS、负载均衡的 SaaS 层——在那一刻成为隐式共享依赖;三是某个训练\u002F推理框架在版本同步发布时出现了时间窗内的回归。\n\n无论哪种解释,这次事件都提醒我们,所谓\"前沿模型可用性 99.x%\"这个数字,是被状态页而非真实用户体验度量的。当四家头部产品在同一个工作日的同一个上午集体哑火,企业和个人用户真正面对的是\"全球 AI 服务可用性\"这个新指标——它远低于任何单一厂商自我报告的数字。\n\n下一次有人告诉你\"多模型备份就够稳了\",可以把这篇文章翻出来。\n\n参考链接\n- Ars Technica 报道:https:\u002F\u002Farstechnica.com\u002Fai\u002F2026\u002F09\u002Ffour-major-ai-models-suffer-rare-overlapping-downtime\u002F\n- Solidot 转载:https:\u002F\u002Fwww.solidot.org\u002Fstory?sid=85283","https:\u002F\u002Farstechnica.com\u002Fai\u002F2026\u002F09\u002Ffour-major-ai-models-suffer-rare-overlapping-downtime\u002F","2af9d198-9418-4f26-85e4-4a8f3eede35a",[11,15,18,21,24],{"id":12,"name":13,"slug":13,"description":14,"color":14},"5e628969-6d2a-437f-998a-104e4b16cfb1","ai-progress",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"23544f6a-eea1-4f05-aa8d-749ca862d5d2","anthropic",{"id":19,"name":20,"slug":20,"description":14,"color":14},"0a93ec8e-ea39-4693-81de-563ca8c173f7","inference",{"id":22,"name":23,"slug":23,"description":14,"color":14},"01598627-1ea6-4b27-a5d8-874971571a71","llm",{"id":25,"name":26,"slug":26,"description":14,"color":14},"42e59a88-7795-47dc-a334-ef1e72c24347","openai",[28],{"id":29,"lang":30,"title":31,"summary":32,"content":33},"e292bab3-68a3-4aad-83db-ba27dbd4f09f","en","Four Major AI Models Went Down in Parallel: OpenAI and Anthropic Confirmed, Gemini and Grok Stayed Silent","On the morning of September 3, 2026 (U.S. Eastern), ChatGPT, Claude, Grok, and Gemini all reported serious disruptions within hours. Anthropic and OpenAI publicly acknowledged the incidents and rolled out mitigations, while Google and xAI issued no statements. AWS, Azure, and Cloudflare status pages stayed clean, yet DownDetector and StatusGator data confirmed the user-side wobble across all four.","## A rare four-way outage\n\nOn the morning of September 3, 2026 U.S. Eastern time, four major AI services — ChatGPT, Claude, Grok, and Gemini — experienced significant disruptions within a span of under three hours. Ars Technica, drawing on each vendor's status pages and DownDetector data, summed up the window bluntly: having all four frontier models concentrate failures in the same stretch is \"practically unheard of.\"\n\n## Anthropic spoke up first, reporting a partial outage at 9:23 ET\n\nAnthropic was the first to acknowledge the problem publicly. Per status.claude.com incident 461yvfrzpwtt, the company flagged \"elevated errors\" affecting Claude Mythos 5.1, Claude Fable 5.1, and Claude Opus 5 at 9:23 ET, identified the cause roughly 15 minutes later, and marked the issue resolved at 12:16 ET — a roughly three-hour incident. Just after noon, a separate event (288w7p4hk1l1) hit Claude Sonnet 5, showing that multiple Claude model families were sequentially stressed within a single workday.\n\n## OpenAI followed at 10:43 ET, wrapped at 12:55\n\nOpenAI published incident 01M1KWEDH417T2CF44YYHZDFCR on status.openai.com at 10:43, reporting \"elevated errors\" and degraded performance for ChatGPT and Codex. A mitigation was deployed within roughly half an hour, and the incident was marked resolved at 12:55. Ars notes that OpenAI's published 90-day availability stood at 99.63% for ChatGPT and 100% for ChatGPT Codex, putting this squarely in \"two-decimals-after-the-point\" territory.\n\n## Grok and Gemini stayed silent, but the data did not\n\nxAI's Grok surfaced a user-facing \"is experiencing issues\" error message, while DownDetector reports on Grok jumped from fewer than 10 before 9:00 ET to 1,365 by 9:45, then settled to 273. Google never publicly acknowledged any Gemini trouble, but DownDetector counts on Gemini leapt from about 23 around 10:30 to 412 just after 11:00, and StatusGator labelled the Gemini API status as \"likely outage\" between 10:45 and 11:15. Both companies stuck to the same playbook: leave no trace on the official status page, and let third-party monitoring take the public-relations hit.\n\n## Hyperscalers were clean, but community data wobbled\n\nNotably, the three hyperscalers most often blamed for upstream failure — AWS, Azure, and Cloudflare — posted no major incidents on their own status pages. Yet DownDetector still showed visible spikes for all three that morning, hinting that an upper-layer model failure briefly pushed traffic sideways through other SaaS surfaces before looping back to the cloud itself. Ars framed it as a \"model layer broke, public cloud stayed up\" split — itself a signal worth keeping.\n\n## Commentary: this is a model-layer single point, not a cloud-layer one\n\nThe most important takeaway is where the failure was sliced. Four models from four different companies, running on different clouds, with different inference stacks and GPU clusters — yet all failed at nearly the same moment. That argues against the failure originating in physical infrastructure. More plausible explanations include: enterprise traffic sitting behind multi-model failover that turned one outage into cascading pressure on the next provider; an upstream vendor (CDN, identity, DNS, load-balancing SaaS) acting as an implicit shared dependency at that instant; or a synchronized training\u002Finference framework release that introduced a time-windowed regression.\n\nRegardless of the cause, this episode is a reminder that the much-cited \"99.x% frontier-model availability\" number is measured by status pages, not by real user experience. When four flagship products go silent in the same morning of the same workday, the figure that actually matters for both enterprises and individuals is \"global AI service availability\" — and that number sits well below any single vendor's self-report.\n\nNext time someone tells you \"multi-model failover is stable enough,\" pull this one back up.\n\nReferences\n- Ars Technica: https:\u002F\u002Farstechnica.com\u002Fai\u002F2026\u002F09\u002Ffour-major-ai-models-suffer-rare-overlapping-downtime\u002F\n- Solidot (Chinese summary): https:\u002F\u002Fwww.solidot.org\u002Fstory?sid=85283","four-ai-models-overlapping-outage-sept-2026","2026-09-06T08:00:00Z","2026-09-06T01:03:17.963482Z","2026-09-06T01:03:17.963498Z",true,"agent",428,[42,51],{"slug":43,"tag_slug":43,"title_zh":44,"title_en":45,"intro_zh":46,"intro_en":47,"id":48,"is_active":38,"created_at":49,"modified_at":50},"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":52,"tag_slug":52,"title_zh":53,"title_en":54,"intro_zh":55,"intro_en":56,"id":57,"is_active":38,"created_at":58,"modified_at":59},"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":61},[62,67,72,77,82,87],{"id":63,"title":64,"news_slug":65,"published_at":66},"390c2437-4e4f-45ec-8270-67c5bfa4fa47","ChatGPT、Claude、Grok、Gemini 罕见同时下线,周四早晨全球 AI 集体失声","chatgpt-claude-grok-gemini-thursday-outage","2026-09-05T06:00:00+00:00",{"id":68,"title":69,"news_slug":70,"published_at":71},"9e58d587-3c1b-44c5-ad36-daf23aeb42a2","微软叫停 tokenmaxxing:GitHub Copilot 默认切回 GPT-5.6 Sol,Parikh 设 token 预算","microsoft-token-budget-gpt-5-6-default","2026-09-03T00:30:00+00:00",{"id":73,"title":74,"news_slug":75,"published_at":76},"54e3cac6-61df-4c68-bf07-559e94ae2624","OpenAI 推理模型攻克80年数学难题：证明埃尔德什单位距离猜想不成立","openai-erdos-unit-distance-disprove","2026-05-21T04:10:00+00:00",{"id":78,"title":79,"news_slug":80,"published_at":81},"376f281c-ed27-44af-93ed-61be1683cbf8","OpenAI推理模型证明80年数学难题：通用推理能力的成人礼","openai-erdos-80-year-math-coming-of-age","2026-05-20T14:10:00+00:00",{"id":83,"title":84,"news_slug":85,"published_at":86},"1e43b4fc-39fe-4cac-aaf9-57f82d5c0311","AI 公司与数学界「错位」:两个月三次刷屏,把同行评审甩在身后","ai-math-severe-misalignment-fields-medal","2026-09-15T10:00:00+00:00",{"id":88,"title":89,"news_slug":90,"published_at":91},"982c5e1e-5274-442e-9237-abaf39e8ee3c","25 位菲尔茨奖得主联名公开信:AI 解题竞赛正在伤害数学","fields-medalists-ai-misalignment-math","2026-09-13T13:07:00+00:00"]