{"day":"2026-09-04","timezone":"Asia/Shanghai","generated_at":"2026-09-04T01:26:01.171634275Z","topics":[{"id":"ai","name":"X","name_en":"AI Industry","description":"大模型、芯片、融资与政策的每日要闻","latest_day":"2026-09-04","updated_at":"2026-09-04T01:05:17.741139Z","total":10,"items":[{"id":"4cad2586-76e1-4e91-8ea1-48a702192efb","title":"英伟达发布PAIR：将家中闲置PC和GPU组成本地AI推理集群","summary":"英伟达推出开源工具Personal AI Router（PAIR），可将家中配备RTX GPU、DGX Spark或Apple M4芯片的闲置电脑组成本地虚拟集群，为多智能体任务智能分配推理负载，兼容Ollama和LM Studio且无需修改现有设置。三设备演示将五子智能体任务时间从18分钟缩短至8分48秒，有助于缓解本地AI瓶颈并保障数据隐私。","title_en":"NVIDIA Launches PAIR to Turn Idle Home PCs and GPUs into Local AI Inference Cluster","summary_en":"NVIDIA released the open-source Personal AI Router (PAIR), which discovers and routes independent inference requests across compatible home-network devices with RTX GPUs, DGX Spark or Apple M4+ silicon, working seamlessly with Ollama and LM Studio without agent changes. A three-device demo cut a five-subagent task from 18 minutes to 8 minutes 48 seconds, easing multi-agent bottlenecks while keeping data local.","url":"https://developer.nvidia.com/blog/nvidia-pair-virtual-inference-router-expands-available-compute-on-your-local-network/","source":"NVIDIA Technical Blog","published_at":"2026-09-03T00:00:00Z","first_seen_at":"2026-09-04T01:05:17.741139Z","tags":["Nvidia","AI硬件","本地推理","智能体","算力"],"importance":4,"pinned":false},{"id":"3e4a9395-1950-48c9-8b13-f6cdb96e0514","title":"沙特HUMAIN发布基于中国MiniMax的阿拉伯语大模型humain-m3","summary":"沙特主权基金PIF支持的AI公司HUMAIN发布国家阿拉伯语大模型humain-m3，基于中国MiniMax实验室的开源M3框架构建，旨在强化阿拉伯AI生态并推动全球前沿智能发展。此举凸显中东国家采用中国开源模型建设主权AI的趋势，与HUMAIN同期宣布的超大规模VC基金形成互补。","title_en":"Saudi HUMAIN Unveils Arabic AI Model humain-m3 Based on China's MiniMax","summary_en":"PIF-backed Saudi AI firm HUMAIN unveiled its national Arabic-language model humain-m3, built on the open-source M3 framework from Chinese lab MiniMax, to strengthen the Arabic AI ecosystem and contribute to global frontier intelligence. The launch highlights Middle Eastern adoption of Chinese open models for sovereign AI, complementing HUMAIN's recently announced giant VC fund.","url":"https://www.bloomberg.com/news/articles/2026-09-03/saudi-arabia-s-humain-unveils-ai-model-based-on-china-s-minimax","source":"Bloomberg","published_at":"2026-09-03T00:00:00Z","first_seen_at":"2026-09-04T01:05:17.741139Z","tags":["沙特","HUMAIN","MiniMax","开源模型","阿拉伯语AI"],"importance":4,"pinned":false},{"id":"7e968d56-770f-45cb-9397-730c9e2887cf","title":"沙特AI公司HUMAIN今年将推出超大规模全球VC基金","summary":"沙特领先AI公司HUMAIN计划今年推出可能超过100亿美元的全球风险投资基金，办公室设在美国、沙特、法国和英国，投资条件包括使用沙特数据中心或引进人才。该公司由公共投资基金（PIF）控制，已成功投资xAI，旨在加速沙特成为全球AI前沿。","title_en":"Saudi AI Firm HUMAIN to Launch Giant Global VC Fund This Year","summary_en":"Saudi Arabia’s leading AI firm HUMAIN will launch a large venture capital fund this year, potentially bigger than a planned $10 billion vehicle, with offices in the US, Saudi Arabia, France and the UK. Backed by PIF, it will invest in AI companies committing to Saudi data centers or bringing staff to the kingdom.","url":"https://www.semafor.com/article/09/03/2026/saudi-ai-firm-humain-to-launch-giant-global-vc-fund-this-year","source":"Semafor","published_at":"2026-09-03T00:00:00Z","first_seen_at":"2026-09-04T00:04:39.659762Z","tags":["HUMAIN","funding","Saudi AI","VC","PIF"],"importance":4,"pinned":false},{"id":"e0b3d61a-0a2a-4307-8943-e86386b7e083","title":"英伟达Nemotron-3-Ultra-CC在IOI 2026上以535.4分超越顶尖人类选手","summary":"英伟达5500亿参数Nemotron-3-Ultra-CC模型在2026年国际信息学奥林匹克（IOI）中得分535.4/600，超过金牌门槛361.12和最高人类选手498.27分，成为首个在IOI问题集上超越顶尖人类的AI系统。该成果通过大规模问题策展、合成推理轨迹、监督微调和测试时计算策略GenCorrect实现。","title_en":"Nvidia's Nemotron-3-Ultra-CC Scores 535.4 at IOI 2026, First AI to Beat Top Human Coder","summary_en":"Nvidia's 550-billion-parameter Nemotron-3-Ultra-CC scored 535.4 out of 600 at IOI 2026, exceeding the gold threshold of 361.12 and the top human contestant's 498.27, marking the first AI system to outscore the highest-scoring human on an IOI problem set under identical constraints.","url":"https://aiweekly.co/alerts/nvidias-550b-nemotron-beats-top-human-coder-at-ioi-2026","source":"AI Weekly","published_at":"2026-09-03T00:00:00Z","first_seen_at":"2026-09-04T00:04:39.659762Z","tags":["Nvidia","Nemotron","IOI","coding","paper"],"importance":4,"pinned":false},{"id":"518dd335-376d-4950-9476-fca237099e3f","title":"Cohere发布Parse 5：多模态模型高效从复杂文档提取结构化信息","summary":"Cohere推出Parse 5，一款23亿参数的多模态基础模型，可将视觉丰富的PDF转换为Markdown并提供边界框坐标以实现视觉接地，在超过2000页企业文档上平均得分79.2。该模型针对复杂企业文档的结构化数据提取进行了优化，提升了AI在信息处理任务中的效率。","title_en":"Cohere Launches Parse 5: Multimodal Model for Efficient Structured Data Extraction from Complex Documents","summary_en":"Cohere has launched Parse 5, a 2.3-billion-parameter multimodal foundation model that converts visually rich PDFs into Markdown with bounding box coordinates for visual grounding. Evaluated against over 2,000 enterprise pages, it achieved an average score of 79.2 in key performance areas.","url":"https://www.infoq.com/news/2026/09/cohere-multimodal-parse/","source":"InfoQ","published_at":"2026-09-03T00:00:00Z","first_seen_at":"2026-09-04T00:04:39.659762Z","tags":["Cohere","multimodal","document AI","Parse 5"],"importance":3,"pinned":false},{"id":"4b21837c-8d9d-418b-b1ec-bbbce3f4e876","title":"OpenAI发布GPT-6 Astra：首个达Critical网络安全等级的模型，称进入AGI时代","summary":"OpenAI发布旗舰模型GPT-6 Astra，自称全球最智能且对齐的模型，首次达到其准备框架中的“Critical”网络安全能力等级；公司总裁Greg Brockman称这是能力代际跃升，或标志AGI时代开启。","title_en":"OpenAI launches GPT-6 Astra, first model rated Critical for cyber and hailed as AGI-era leap","summary_en":"OpenAI launched flagship GPT-6 Astra, calling it the world’s most intelligent and aligned model and the first to reach the Critical cybersecurity tier of its Preparedness Framework; president Greg Brockman described a generational leap that may mark the start of the AGI era.","url":"https://openai.com/index/responding-next-frontier-critical-cyber-capabilities/","source":"OpenAI","published_at":"2026-09-03T00:00:00Z","first_seen_at":"2026-09-03T22:01:07.137639Z","tags":["OpenAI","GPT-6 Astra","大模型发布","网络安全","AGI"],"importance":5,"pinned":false},{"id":"f395c0e3-a3cf-4118-9fb6-912bee4531e9","title":"英伟达129亿美元收购Hugging Face，开源AI平台将保持开放","summary":"英伟达宣布以129.3亿美元收购Hugging Face，这是其史上最大收购案。Hugging Face将继续作为开放平台运营，支持所有模型和硬件，这对开源AI生态系统至关重要，可能加速模型分发并巩固英伟达在AI基础设施中的核心地位。","title_en":"NVIDIA to Acquire Hugging Face for $12.93 Billion, Open AI Platform to Remain Open","summary_en":"NVIDIA announced a $12.93 billion acquisition of Hugging Face, its largest deal ever. Hugging Face will remain an open platform supporting all models and hardware, a pivotal move for the open-source AI ecosystem that could accelerate model distribution and reinforce NVIDIA's central role in AI infrastructure.","url":"https://blogs.nvidia.com/blog/nvidia-to-acquire-hugging-face/","source":"NVIDIA Blog","published_at":"2026-09-03T00:00:00Z","first_seen_at":"2026-09-03T20:52:13.777413Z","tags":["英伟达","Hugging Face","收购","开源AI","大模型"],"importance":5,"pinned":false},{"id":"b7ae83eb-b97e-4507-936a-685a44f617b8","title":"Google DeepMind发布WeatherNext 3：每小时刷新、5公里分辨率的全球天气AI模型","summary":"Google DeepMind推出最先进的全球天气预测AI模型WeatherNext 3，支持每小时更新和高达5公里分辨率，并整合实时卫星数据，显著提升降水和可再生能源预测准确性，已集成到Google搜索、地图等产品。","title_en":"Google DeepMind Unveils WeatherNext 3: Hourly, 5km-Resolution Global Weather AI Model","summary_en":"Google DeepMind launched WeatherNext 3, its most advanced global weather AI model with hourly updates and up to 5km resolution, incorporating real-time satellite data for improved precipitation and renewable energy forecasts. It is now integrated into Google Search, Maps, and other products.","url":"https://blog.google/innovation-and-ai/models-and-research/google-deepmind/introducing-weathernext-3/","source":"Google Blog","published_at":"2026-09-03T00:00:00Z","first_seen_at":"2026-09-03T20:52:13.777413Z","tags":["DeepMind","天气AI","WeatherNext","Google","科学AI"],"importance":4,"pinned":false},{"id":"71237373-8f2f-4434-98f6-e0e5831a46fa","title":"四大主流AI模型罕见同时宕机，ChatGPT、Claude、Grok和Gemini受影响","summary":"OpenAI、Anthropic、xAI和Google的主要AI模型在周四上午出现罕见重叠服务中断，持续数小时。这凸显了云端AI基础设施的潜在脆弱性，多数服务已恢复，但Grok仍有问题。","title_en":"Four Major AI Models Suffer Rare Overlapping Downtime Hitting ChatGPT, Claude, Grok and Gemini","summary_en":"Major AI models from OpenAI, Anthropic, xAI, and Google experienced a rare overlapping outage lasting several hours on Thursday morning. This highlights potential vulnerabilities in cloud AI infrastructure; most services have recovered, though Grok continues to have issues.","url":"https://arstechnica.com/ai/2026/09/four-major-ai-models-suffer-rare-overlapping-downtime/","source":"Ars Technica","published_at":"2026-09-03T00:00:00Z","first_seen_at":"2026-09-03T20:52:13.777413Z","tags":["AI宕机","ChatGPT","Claude","Grok","Gemini"],"importance":4,"pinned":false},{"id":"ad52b617-c625-4a28-9e5a-49ee5fd4266d","title":"Figure与Nscale达成35亿美元算力合作，部署10万颗NVIDIA GPU训练人形机器人","summary":"人形机器人公司Figure与云服务商Nscale签署战略合作，计划部署高达10万颗NVIDIA Vera Rubin GPU，初始承诺35亿美元算力，用于训练Helix模型，加速通用机器人智能发展。","title_en":"Figure and Nscale Sign $3.5 Billion Compute Deal to Deploy 100,000 NVIDIA GPUs for Humanoid Training","summary_en":"Humanoid robot company Figure signed a strategic partnership with Nscale to deploy up to 100,000 NVIDIA Vera Rubin GPUs, with an initial $3.5 billion compute commitment to train its Helix model and accelerate general-purpose robotics intelligence.","url":"https://www.figure.ai/news/figure-and-nscale-sign-strategic-partnership","source":"Figure AI","published_at":"2026-09-03T00:00:00Z","first_seen_at":"2026-09-03T20:52:13.777413Z","tags":["人形机器人","Figure","NVIDIA","算力","AI硬件"],"importance":4,"pinned":false}]},{"id":"openai","name":"OpenAI","name_en":"OpenAI","description":"OpenAI 官方动态与最新模型","latest_day":"2026-09-04","updated_at":"2026-09-04T01:23:37.612821Z","total":13,"items":[{"id":"1cc9954f-d4ba-46e3-a8d5-56c346ba0273","title":"Sam Altman 为 GPT-6 Astra 混乱推出道歉，计划很快向 Pro 和 API 用户广泛开放","summary":"OpenAI CEO Sam Altman 引用补偿公告，为 GPT-6 Astra 的混乱推出道歉，表示公司会尽力纠正，并计划在不久的将来开始向 API 客户和 ChatGPT 订阅者广泛推出，首先从 Pro 用户开始。他希望用户本周末就能使用，但尚无法承诺。","title_en":"Sam Altman Apologizes for Messy GPT-6 Astra Rollout, Plans Broad Access for Pro and API Users Soon","summary_en":"OpenAI CEO Sam Altman apologized for the messy GPT-6 Astra rollout and said the company tries to make it right. Broad access for API customers and ChatGPT subscribers is expected in the near future, starting with Pro users. He is hopeful people can use it this weekend but cannot promise yet.","url":"https://x.com/sama/status/2095678759651438887","source":"Sam Altman (@sama)","published_at":"2026-09-04T01:02:06Z","first_seen_at":"2026-09-04T01:23:37.612821Z","tags":["GPT-6 Astra","推出","Sam Altman","访问","Pro"],"importance":3,"pinned":false},{"id":"592e29ca-356b-4d5f-9acd-083419f743a2","title":"Codex CLI 0.153.2 发布，修正 GPT-6 Astra Fast 档描述为 2 倍速度","summary":"OpenAI 发布 Codex CLI 0.153.2，将 GPT-6-Astra Fast 档描述从“1.5x”更正为“2x speed, increased usage”。此更新仅修改显示文本，不影响实际请求行为，为开发者提供更准确的 Astra 使用信息。","title_en":"Codex CLI 0.153.2 Released, Corrects GPT-6 Astra Fast Tier Description to 2x Speed","summary_en":"OpenAI released Codex CLI 0.153.2, correcting the GPT-6-Astra Fast tier description from “1.5x” to “2x speed, increased usage.” The change affects displayed text only and does not alter request behavior, giving developers more accurate information on Astra usage.","url":"https://github.com/openai/codex/releases/tag/rust-v0.153.2","source":"GitHub","published_at":"2026-09-03T23:53:00Z","first_seen_at":"2026-09-04T00:22:47.985361Z","tags":["Codex CLI","GPT-6 Astra","更新","开发者工具"],"importance":2,"pinned":false},{"id":"66e07cdd-2e59-4009-9cde-cf029775e144","title":"Tibo：付费用户因 Astra 延迟访问将获每日银行重置补偿","summary":"OpenAI 高管 Tibo 宣布，自今日起，付费 ChatGPT 计划用户每延迟一天无法使用 GPT-6 Astra，将获得一次银行用量重置作为补偿。团队正全力加速推出，首次重置预计约 3 小时后到账。","title_en":"Tibo: Paid users to get banked resets for each day without Astra access","summary_en":"OpenAI exec Tibo announced that starting today, paid ChatGPT plan users will receive one banked usage reset for every day they lack access to GPT-6 Astra. The team is rushing the rollout, with the first reset expected in about 3 hours.","url":"https://x.com/thsottiaux/status/2095651088502591861","source":"X / @thsottiaux","published_at":"2026-09-03T23:12:09Z","first_seen_at":"2026-09-03T23:22:27.140166Z","tags":["GPT-6 Astra","用量补偿","ChatGPT","推出"],"importance":4,"pinned":false},{"id":"5c3025f3-2a11-41d2-8d98-0ff81dbd86c6","title":"OpenAI API 为 GPT-6 Astra 新增异步工具调用、中途转向和推理调整功能","summary":"OpenAI API Changelog 宣布 GPT-6 Astra 上线 Responses 和 Chat Completions 接口，同时新增长时任务控制：异步工具调用（模型可在应用执行工具时继续推理）、通过 WebSocket 的中途指令转向，以及对话中途调整推理力度。这些功能专为复杂代理工作流设计。","title_en":"OpenAI API Adds Async Tool Calling, Mid-Turn Steering and Reasoning Controls for GPT-6 Astra","summary_en":"The OpenAI API changelog released GPT-6 Astra on Responses and Chat Completions endpoints, plus new controls for long-running work: async tool calling so the model continues while tools execute, mid-turn steering over WebSockets, and changing reasoning effort mid-conversation. These support advanced agentic workflows.","url":"https://developers.openai.com/api/docs/changelog","source":"OpenAI Developers","published_at":"2026-09-03T22:13:00Z","first_seen_at":"2026-09-03T22:23:23.231153Z","tags":["GPT-6 Astra","API","Responses API","异步工具调用"],"importance":4,"pinned":false},{"id":"74c8155a-784b-4a18-8a96-7bc0073fd413","title":"Codex CLI 0.153.1 发布，支持通过 API 配置 GPT-6 Astra","summary":"Codex CLI 发布 0.153.1 版本，新增通过 API 配置 GPT-6 Astra 的支持，无需更改默认模型，也不会在模型选择器中显示。这是配合 GPT-6 Astra 推出的配套更新。","title_en":"Codex CLI 0.153.1 Released with GPT-6 Astra API Configuration Support","summary_en":"Codex CLI 0.153.1 has been released, adding support for configuring GPT-6 Astra via the API without changing the default model or showing it in the model picker. This is a backport to support the new model launch.","url":"https://github.com/openai/codex/releases/tag/rust-v0.153.1","source":"OpenAI Codex / GitHub","published_at":"2026-09-03T21:08:58Z","first_seen_at":"2026-09-03T21:21:00.675632Z","tags":["Codex CLI","GPT-6 Astra","开发者工具","API"],"importance":3,"pinned":false},{"id":"53971bdf-2fd7-42eb-a788-f8e89e3d990f","title":"Sam Altman：正尽快让所有人用上 GPT-6 Astra，感谢用户耐心","summary":"OpenAI CEO Sam Altman 表示团队正尽快将 Astra 交到所有用户手中，理解当前滚动推出可能令人沮丧，感谢耐心，并预计很快完成。他同时转发官方视频并称这是迄今最喜欢的 OpenAI 视频，对未来感到兴奋。","title_en":"Sam Altman: Working to Get GPT-6 Astra to Everyone Quickly, Thanks for Patience","summary_en":"OpenAI CEO Sam Altman said the team is working to get Astra into everyone's hands as quickly as possible, acknowledging frustration with the staged rollout and thanking users for patience. He expects it to be swift and called the official launch video his favorite OpenAI video yet.","url":"https://x.com/sama/status/2095601211869421726","source":"X/@sama","published_at":"2026-09-03T19:53:57Z","first_seen_at":"2026-09-03T22:23:23.231153Z","tags":["Sam Altman","GPT-6 Astra","推出进度"],"importance":3,"pinned":false},{"id":"23befe6b-49fc-4691-8ee4-bcce9b48e1d2","title":"Tibo：GPT-6 Astra 向所有 Plus 用户推出，纳入常规用量且可 100% 使用","summary":"OpenAI Codex与ChatGPT负责人Tibo发帖称，公司正谨慎且快速地推出GPT-6 Astra，特别强调将覆盖所有Plus用户而非仅限Pro、Business和Enterprise。推出预计需几天，背后首次大规模运行多项新系统。随后确认Astra将纳入各计划常规用量额度，用户可将100%配额用于该模型。","title_en":"Tibo: GPT-6 Astra Rolling Out to All Plus Users, Included in Standard Quotas at 100%","summary_en":"Tibo (@thsottiaux), OpenAI's Codex and ChatGPT lead, posted that GPT-6 Astra is being released carefully and quickly to all Plus users, not just higher tiers. The rollout will take a few days as novel systems scale for the first time. Astra is included in standard usage allocations, allowing 100% of quota toward it.","url":"https://x.com/thsottiaux/status/2095597168816226335","source":"X/@thsottiaux","published_at":"2026-09-03T19:37:54Z","first_seen_at":"2026-09-03T22:23:23.231153Z","tags":["GPT-6 Astra","ChatGPT Plus","用量额度","Tibo"],"importance":4,"pinned":false},{"id":"ead1c7c6-165f-4837-a1bf-e2704953e407","title":"OpenAI 开发者账号展示 GPT-6 Astra 编程、计算机使用和 UI 设计实测表现","summary":"@OpenAIDevs 发布开发者向帖文，展示 GPT-6 Astra 在 DeepSWE v1.1 编程基准达 75.2%、OSWorld 2.0 计算机使用达 72.6%，并具备将草图转化为可工作 UI、优化布局与交互的视觉判断能力。模型即将登陆开发者构建环境。","title_en":"OpenAI Developers Account Details GPT-6 Astra Coding, Computer-Use and UI Benchmarks","summary_en":"@OpenAIDevs posted developer-focused details showing GPT-6 Astra scoring 75.2% on DeepSWE v1.1 for coding and 72.6% on OSWorld 2.0 for computer use, plus strong visual capabilities to turn sketches into functional UIs with layout and interaction refinements. Availability for builders is coming very soon.","url":"https://x.com/OpenAIDevs/status/2095596178117419365","source":"X/@OpenAIDevs","published_at":"2026-09-03T19:33:57Z","first_seen_at":"2026-09-03T22:23:23.231153Z","tags":["GPT-6 Astra","开发者","编程基准","计算机使用"],"importance":3,"pinned":false},{"id":"b5f75e05-3d65-4640-af6d-fbc61959e3de","title":"OpenAI 发布 GPT-6 Astra：全球最智能对齐模型，今日开始向 Plus 用户滚动推出","summary":"OpenAI 今日发布 GPT-6 Astra，宣称是全球最智能且对齐的模型，在 FrontierMath Tier 4（98%）、ARC-AGI 3（99.9%）、计算机使用、软件工程、科学发现和网络安全等多项基准上达到新 SOTA。该模型今日向有限组织滚动推出，未来几天将向所有 ChatGPT Plus、Pro、Business 和 Enterprise 用户以及 OpenAI API 和 AWS 开放。","title_en":"OpenAI Launches GPT-6 Astra, the World's Most Intelligent and Aligned Model, Rolling Out to Plus Users","summary_en":"OpenAI launched GPT-6 Astra today, calling it the world's most intelligent and aligned model with new SOTA results on FrontierMath Tier 4 (98%), ARC-AGI 3 (99.9%), computer use, software engineering, science, and cybersecurity. It is rolling out now to a limited set of organizations and will become available to all ChatGPT Plus, Pro, Business, and Enterprise users as well as the OpenAI API and AWS over the coming days.","url":"https://openai.com/index/gpt-6-astra/","source":"OpenAI","published_at":"2026-09-03T19:32:13Z","first_seen_at":"2026-09-03T21:21:00.675632Z","tags":["GPT-6 Astra","模型发布","ChatGPT","计算机使用","API"],"importance":5,"pinned":false},{"id":"4462b9f9-3449-4e54-9930-7b6947cec14a","title":"OpenAI 推出 Daybreak for Frontline Defenders：承诺 10 亿美元助力一线网络安全防御","summary":"OpenAI 宣布「Daybreak for Frontline Defenders」倡议，承诺投入 10 亿美元补贴 Daybreak 网络模型访问、培训和技术支持，优先帮助水电、地方政府、社区银行等资源有限的一线防御者保护关键服务。该计划与 GPT-6 Astra 达到关键网络安全能力同步推出，旨在抓住防御窗口、加速漏洞发现与修复。","title_en":"OpenAI Launches Daybreak for Frontline Defenders with $1B Commitment to Protect Essential Services","summary_en":"OpenAI launched Daybreak for Frontline Defenders, committing $1 billion in subsidized access, training and support so resource-constrained defenders—including water utilities, local governments and community banks—can use frontier AI to protect essential services. The initiative coincides with GPT-6 Astra reaching the Critical cybersecurity threshold and aims to close the defender’s window before AI-enabled attacks scale.","url":"https://openai.com/index/daybreak-for-frontline-defenders/","source":"OpenAI","published_at":"2026-09-03T00:00:00Z","first_seen_at":"2026-09-04T01:23:37.612821Z","tags":["Daybreak","网络安全","$1B承诺","关键基础设施","GPT-6 Astra"],"importance":4,"pinned":false},{"id":"c6221736-c5f8-4435-90b5-a6e39d32119c","title":"GPT-6 Astra 现已通过 Microsoft Foundry 有限访问计划推出","summary":"微软宣布 GPT-6 Astra 即日起通过 Microsoft Foundry 有限访问计划向企业客户推出，支持跨应用的复杂工作流执行、计算机使用及专业产出，并将在未来数日扩大可用性。","title_en":"GPT-6 Astra now available in Microsoft Foundry Limited Access Program","summary_en":"Microsoft announced that GPT-6 Astra is rolling out today via the Microsoft Foundry Limited Access Program for enterprise customers, enabling agentic workflows, computer use, and professional outputs, with broader availability in the coming days.","url":"https://azure.microsoft.com/en-us/blog/gpt-6-astra-frontier-intelligence-for-work-now-available-in-microsoft-foundry/","source":"Microsoft Azure Blog","published_at":"2026-09-03T00:00:00Z","first_seen_at":"2026-09-03T23:22:27.140166Z","tags":["GPT-6 Astra","Azure","Foundry","企业","合作"],"importance":3,"pinned":false},{"id":"b21d353c-81c7-482d-872d-b641c8114062","title":"OpenAI 发布 GPT-6 Astra 安全概览：达关键网络安全能力水平","summary":"OpenAI 发布 GPT-6 Astra 安全概览，确认该模型为首个达到准备框架关键网络安全能力阈值的模型，并详述了针对误用、错位及越狱的强化防护与对齐改进。","title_en":"OpenAI publishes GPT-6 Astra safety overview: reaches Critical cyber capability","summary_en":"OpenAI published a safety overview for GPT-6 Astra, stating it is the first model to reach the Critical cybersecurity capability threshold under its Preparedness Framework, with strengthened safeguards against misuse, misalignment, and jailbreaks.","url":"https://openai.com/index/safety-overview-gpt-6-astra/","source":"OpenAI","published_at":"2026-09-03T00:00:00Z","first_seen_at":"2026-09-03T23:22:27.140166Z","tags":["GPT-6 Astra","安全","网络安全","对齐"],"importance":4,"pinned":false},{"id":"60a2569a-1dd2-4879-b29c-990db6bc28e1","title":"ChatGPT 发布说明：GPT-6 Astra 新增安全监控，必要时可暂停对话供审查","summary":"ChatGPT 发布说明更新介绍 GPT-6 Astra，强调其在编码、研究、计算机使用和多步工作上的提升，并新增安全监控：若检测到代理可能未正确理解指令，对话可能被暂停或停止，供用户审查后决定如何继续。目前仅限有限组织，更广泛推出即将开始。","title_en":"ChatGPT Release Notes: GPT-6 Astra Adds Safety Monitoring That May Pause Conversations","summary_en":"ChatGPT release notes introduce GPT-6 Astra with gains in coding, research, computer use and multi-step work, plus new safety monitoring that may pause or stop a conversation if an agent appears to have misinterpreted instructions, allowing user review. Access remains limited for now, with broader availability coming.","url":"https://help.openai.com/articles/6825453-chatgpt-release-notes","source":"OpenAI Help Center","published_at":"2026-09-03T00:00:00Z","first_seen_at":"2026-09-03T22:23:23.231153Z","tags":["ChatGPT","GPT-6 Astra","安全监控","发布说明"],"importance":3,"pinned":false}]},{"id":"claude","name":"Claude","name_en":"Claude","description":"Anthropic 与 Claude 的官方动态与最新模型","latest_day":"2026-09-04","updated_at":"2026-09-03T23:20:38.133372Z","total":1,"items":[{"id":"4b6a04e2-67b1-45c8-9a86-c10c0efeb0be","title":"Anthropic 发布 Claude 最新官方动态","summary":"Anthropic 发布 Claude 最新能力更新与开发者工具进展，官方账号与博客同步披露模型、API 及 Claude Code 相关动态。","title_en":"Anthropic publishes latest official Claude updates","summary_en":"Anthropic released the latest Claude capability updates and developer tooling progress, with official accounts and the blog disclosing model, API, and Claude Code developments.","url":"https://www.anthropic.com/news","source":"Anthropic","published_at":"2026-09-03T16:00:00Z","first_seen_at":"2026-09-03T23:20:38.133372Z","tags":["Anthropic","Claude","API"],"importance":4,"pinned":false}]}]}