聚合官方博客 · 权威媒体 · 深度通讯 · 学术信号。
只做索引，不做转载，标题与链接均指向原始发布方。
📝 今日导语 从 Opus 5.5 的长任务成本，到 MiMo 强化学习与美团统一推荐，判断模型如何进入真实流程。 🏷 今日关键词： Claude Opus 5.5 MiMo V2.6 任务级评估 Agent Harness 企业上下文 📰 今日精选内容 Anthropic 发布 Claude Opus 5.5，长任务成本降四成 Hacker News · 文章 · 评分 93 Xiaomi MiMo-V2.6：扩展强化学习，迈向自我提
OpenAI推出GPT-6 Sol与GPT-6 Luna 手机新浪网
Founders shouldn't have to learn the hardest lessons the hardest way. TechCrunch Founder Summit is designed to make the challenges of starting a company easier and the highs that much greater.
OpenAI正式发布GPT-6 Sol/Luna！API价格大降50% 免费用户也能用 驱动之家
Now that Paramount has reached a settlement with the 12 states that were suing to block its $110 billion merger with Warner Bros. Discovery (WBD), the studio is even closer to becoming one of the world's biggest producti
HN ⭐ 14 · 2 评论
The seven-year-old startup has raised a $350 million Series E to fuel its data-as-a-service approach.
9月23日外盘头条：特朗普证实美伊在联大罕见直接接触 Anthropic和OpenAI推出成本更低的新模型 新浪财经
为什么Jev必须诞生在OpenAI之外：System One模型与RLHF的隐藏代价-模型 至顶网
The frontier AI model race has entered its comparison shopping phase.
📌 One-Sentence Summary 约翰·普拉特，AI 和科学领域的知名专家，讨论了他在谷歌的 Empirical Research Assistance（ERA）中的工作及其在解决科学问题方面的应用，包括气候变化。 📝 Summary 约翰·普拉特，AI 和科学领域的先驱，分享了他对谷歌的 Empirical Research Assistance（ERA）及其潜力在革命化科学问题解决方面的见解。他强调了 ERA 的能力在自
📌 One-Sentence Summary OpenAI 为 GPT-6 推出改进的提示缓存系统，默认提升缓存命中率，并提供仪表盘、诊断工具和显式缓存断点等控制手段，帮助持久化 Agent 更快更省地运行。 📝 Summary OpenAI 发布 GPT-6 的改进提示缓存系统，面向需要长时间运行的持久化 Agent。新系统默认提供更高的缓存命中率，对 30 分钟内复用的合格共享前缀给予缓存折扣，缓存输入 token 最高可享 90%
Learn how GPT-6 improves prompt caching with higher cache hit rates, new diagnostics, explicit breakpoints, and controls that reduce latency and costs.
OpenAI内部曝光：AI开始“自己造AI” 文学城
Rabbit, the company behind the underwhelming R1 device, is rolling out a standalone AI agent that you don't need its hardware to use, as reported earlier by Wired. The startup says its new OS3 "agentic operating system"
AI ：马克龙呼吁避免依赖美国或中国 RFI
AI也搞“同行评审”？曝OpenAI与Anthropic曾讨论互测彼此模型 新浪财经
📌 One-Sentence Summary Peter Yang 将 Opus 5.5 与 GPT 6 Sol 和 ChatGPT 进行对比，并指出其优劣。 📝 Summary Peter Yang 将 Opus 5.5 与 GPT 6 Sol 和 ChatGPT 进行对比，阐述了对比的背景，并指出其优劣。 📊 Article Meta AI Screening： 92 Source： Peter Yang(@petergyang)
The Kingdom of Saudi Arabia is mostly known for its global dominance over petroleum production and oil reserves - not necessarily cars and auto manufacturing, and certainly not electric vehicle production. So it was a li
OpenAI再推两款GPT-6模型：Sol、Luna定价降50%，Astra能力加速下沉 华尔街见闻
With GPT-6 Sol and Luna, OpenAI adds two cheaper models that deliver their predecessors' performance at half the token price and take aim at Anthropic's pricier offerings. Independent analyses find little gain in actual
OpenAI 发布 GPT-6 Sol 和 Luna 模型，API 价格大降 50% 新浪网
Motorola is the first company to announce a phone running Qualcomm's top-end Snapdragon 8 Elite Extreme Gen 6 chip. The Signature 27 looks set to be Motorola's most advanced flagship in some years, though we're still wai
Qualcomm has announced the Snapdragon 8 Elite Gen 6, this year joined by the 8 Elite Extreme Gen 6 too. The company describes both new phone chips as flagships, and the spec differences are relatively minor, mostly limit
Qualcomm said that its new top chip can run 30B mixture-of-expert model locally.
OpenAI发布GPT-6 Sol和Luna模型，API价格大降50% 搜狐网
Apple's new texture and grain controls for stylizing photos will be more limited on older phones than initially expected. A confusingly worded press release from Apple indicated that the new features, announced alongside
EvilTokens provided an end-to-end platform that makes mass compromises faster and easier.
The Android-powered Googlebooks are coming soon, with familiar brands and specs that rival Windows Copilot machines. For anyone eager to get their hands on any the new laptops, there’s a preorder perk you may want to kno
British Columbia sues OpenAI, demands Tumbler Ridge shooter’s ChatGPT logs.
OpenAI将让第三方机构更早介入AI模型安全评估 新浪财经
Anthropic和OpenAI在呼吁放缓开发之后推出成本更低的新模型 新浪财经
Meta says Muse was built from scratch, but acknowledges the AI assistant was "heavily inspired" by OpenClaw — down to some of its workspace filenames and content.
Anthropic has released Claude Opus 5.5, the first model in its new Claude 5.5 family. The team states it performs at the level of Claude Fable 5.1 on most work. It also costs 40% less to run than Opus 5 on typical worklo
Anthropic发布Claude Opus 5.5模型：性能媲美Fable 5.1，运行成本比Opus 5低40% 凤凰网科技
Apple may pay out up to $95 for each eligible iPhone purchased by someone who felt misled about Siri’s release. You have until December 21 to submit a claim.
HN ⭐ 114 · 66 评论
📌 One-Sentence Summary OpenAI 推出 GPT-6 Sol 和 GPT-6 Luna，这些模型基于 GPT-6 Astra 的进展，提供更快、更便宜的模型来支持大规模工作 📝 Summary GPT-6 Sol 和 GPT-6 Luna 是基于 GPT-6 Astra 的进展而开发的更快、更便宜的模型。它们提供了 50% 的 API 价格降低，相比于 GPT-5.6 的促销价格 📊 Article Meta A
Until the end of the day on September 22nd, 2026, Best Buy and Amazon have the black Apple Magic Keyboard on sale for $169.99, a $30 discount from the usual price. This is the souped-up version of Apple’s full-size, wire
GPT-6 Sol and GPT-6 Luna are now generally available on Amazon Bedrock, giving you more options to match intelligence and efficiency to each workload.
Hey, you know it's like super obvious if you're using AI to write your scripts for TikTok and YouTube, right? [...] It's not just the general AI-isms of "it's not X, it's Y", or the rule of three, or the really weird bro
📌 One-Sentence Summary Peter Yang 分享了一段由 Opus 5.5 生成的金门大桥 3D 场景，并将其与 Astra 进行对比。 📝 Summary Peter Yang 分享了一段由 Opus 5.5 生成的金门大桥 3D 场景，强调其与 Astra 的相似度，并介绍了该场景的创作背景。 📊 Article Meta AI Screening： 90 Source： Peter Yang(@peterg
Meet GPT-6 Sol and Luna, two models that bring frontier intelligence to everyday work with different balances of capability and cost.
OpenAI is launching two new models, which it says are cut from the same cloth as Astra.
Anthropic推出Claude Opus 5.5 宣称能以更低成本处理企业任务 新浪财经
San Francisco is suing the Trump Media & Technology Group, the parent company behind Truth Social, for selling early access to President Donald Trump's posts on the social media site. The lawsuit, filed today in Californ
Claude Opus 5.5, Anthropic's most capable Opus model for agentic coding, knowledge work, and long-running tasks, is now available on Amazon Bedrock and Claude Platform on AWS. This post covers what's new in Opus 5.5, pra
HN ⭐ 41 · 21 评论
Skills let you encode domain-specific procedures as reusable, portable instructions for agents, but a fluent answer doesn't prove the agent picked the right skill or followed it. Learn how to measure skill selection and
Anthropic is launching Claude Opus 5.5, the first model in a new generation. The company says it matches Claude Fable 5.1 on most tasks while costing about 40 percent less to run than Opus 5. Anthropic's benchmarks also
OPENAI将允许外部团体在更早阶段评估AI模型 东方财富
Toyota's push comes as automakers race to develop and deploy humanoid robots.
Apple is working on a new fitness tracker with a screenless design, similar to the one made by Whoop, according to a report from Bloomberg. A prototype of the device reportedly has "a thin fabric band with a sensor-equip
HN ⭐ 215 · 62 评论
软件企业AI产品降价，留住客户以防转投Anthropic、OpenAI 新浪财经
Anthropic 也要亲手做实验了，AI 制药的分水岭到了 搜狐网
Anthropic呼吁放慢开发后首次上新：Claude Opus 5.5能力接近Fable 5.1、成本降四成 华尔街见闻
Anthropic called it "the strongest-performing model we've tested to date."
HN ⭐ 1102 · 769 评论
Two years after trying to sidestep mobile apps with dedicated AI hardware, Rabbit is launching OS3, a cross-platform agent that lives on the screens you already use.
阿里：三年之内会出现一个原生的全模态统一生成模型，未来体验将不再受限于模态边界。
Release: llm-typesafe 0.1a0 I built this new plugin for LLM to add support for TypeSafe AI's new Jev model . Install it like this: llm install llm-typesafe Then set an API key ( get one here , the waitlist seems to move
CPU与GPU趋近1∶1
Reactiv used Amazon Bedrock AgentCore to build a multi-agent AI Scheduler that autonomously refreshes Shopify merchants' mobile apps on a schedule, reducing merchant configuration time by 80% and getting to production 33
Opus 5.5与GPT-6新品同日亮相，AI模型价格战愈演愈烈 财联社
Concurrency sweeps help you right-size a generative AI endpoint on Amazon SageMaker AI by systematically benchmarking it at increasing load levels. This post walks through deploying a model, running automated concurrency
In about four weeks, Trane Technologies built an AI-powered agentic solution on Amazon Bedrock AgentCore that reduced a 20-minute, multi-screen building diagnostic workflow to a 20-second natural language interaction, a
OpenAI wants international standards for recursive self-improvement, where AI systems independently build the next generation of AI. Without safeguards, humans could lose control over this process. OpenAI says the US sho
Learn how Tata Elxsi built IRIS, a real-time industrial safety platform on AWS. IRIS filters camera video at the edge, streams metadata through Amazon Kinesis, runs computer vision on Amazon SageMaker AI, and correlates
Public sector agencies process large volumes of unstructured evidence, such as body camera footage and scanned documents. This post shows how to combine Amazon Bedrock Data Automation with the Model Context Protocol (MCP
软件企业降价推销AI产品，防止客户转投Anthropic、OpenAI 新浪财经
Have you ever read a paper in Science or Nature and thought, “Man, that research was so cool. I wish I could try that method on my own data,” only to spend a week wrestling with someone else’s undocumented repo, broken d
At TechCrunch Disrupt 2026, five sessions across the AI Stage and Real World AI Stage cover AI safety, featuring leaders from Anthropic, Nvidia, AWS, Waabi, and more. Register before September 25 to save up to $200.
Autonomy-1 will have a small, transformer-based AI model taking charge of a space probe.
HN ⭐ 249 · 184 评论
一场两小时的客户拜访，双方说的话加起来上万字。但回到工位，销售能凭记忆敲进 CRM的，通常不到三百字。 剩下的去哪了？跟着出租车一起走了。 这样的对话在企业里每天都在发生。会议室里的争论、客户办公室里随口的一句提醒、老师傅肚里的经验，信息密度最高的这些话，几乎都发生在电脑屏幕之外，从来没有进入过任何一套企业系统。 另一边，已经沉淀在文档、群聊和系统里的数据，也常常因为没做结构化，或者 体量太大，塞不进 上下文 窗口 ， Agent根本调
9月22日，中国高端新能源品牌旗舰车型极氪9X，正式发运阿联酋，开启规模化出海。作为中国新一代豪华旗舰，极氪9X在阿联酋售价超80万元，欧洲售价近100万元人民币，被外媒称为“最贵中国车”，正加速拓展至中东、中亚、欧洲、拉美等核心市场。 极氪9X搭载全球领先的SEP浩瀚超级电混架构，以极致的性能操控和安全能力，在海外市场“未发先火”，部分国家和地区加价超过50万，与保时捷、劳斯莱斯、宾利等海外传统超豪华品牌展开直面竞争。
Hello Robot CEO and co-founder Aaron Edsinger will bring Stretch 4 for a live demo on the Real World AI Stage at TechCrunch Disrupt 2026. Register before September 25 to save up to $200, plus get a second pass at 50% off
从真实工厂到产品复制，再到生态和下一批机会，两位嘉宾会把工业AI从「能用」走向「规模化」的关键问题一层层拆开。
We have reopened our exhibitor program for 1 more week. Book your exhibit table by September 30 at 11:59 p.m. PT and showcase your startup in front of 10,000+ founders, investors, and tech leaders at SF's Moscone West fr
Save up to $200 on your TechCrunch Disrupt 2026 pass, plus 50% off a second pass before prices increase on September 25 at 11:59 p.m. PT. Register today.
The Biological Computing Co. wants to team up with AWS to sell a text-to-video model that's supposed to run five times faster and 80 percent cheaper thanks to a software layer derived from real nerve cells. The tiny laye
商汤U1 Pro，正式发布
HN ⭐ 535 · 354 评论
Anthropic发布AI滥用报告后 中国监管部门据报对DeepSeek和月之暗面展开调查 RFI
The US has spent billions building a “virtual wall” of surveillance towers along its southern border over the past 25 years, promising they will help detect and apprehend border crossers and save lives. But a groundbreak
📌 One-Sentence Summary 与 Epoch AI 的 JS Denain 一起 辩论 RSI 、 美国-中国差距 与 波动性 📝 Summary 本文讨论了 RSI （递归自我改进）的现状及其对 AI 发展的影响，包括美中差距与波动性，并采访了 Epoch AI 的 JS Denain 他分享了对此主题的见解。 💡 Main Points RSI 是现代 AI 战略的关键组成部分，推动创新与竞争 文章强调 RSI 在推
Dyson admits a component issue but won't say what's failing in its $500 toothbrush.
ChatGPT版Grok Bot代码曝光！OpenAI也要为你造不下班的AI同事 华尔街见闻
📌 One-Sentence Summary Jev 是一款颠覆了传统 LLM 架构的系统一模型，它在声明的输出空间上返回概率质量，适用于需要快速且准确做出决策的应用场景。 📝 Summary 本文阐述了 Jev 的架构与特性，这是一款与传统 LLM 不同的系统一模型。文章讨论了 Jev 如何在声明的输出空间上返回概率质量，以及这如何实现更快速、更准确的决策。同时，文章还介绍了适用于 Jev 的五种模式，包括网关 (gate)、路由器
HN ⭐ 53 · 7 评论
HN ⭐ 36 · 8 评论
The Biological Computing Company is bringing its AI tools to Amazon Web Services in a major boost for a once-fringe field that aims to marry nature with code.
No cameras? No problem. Viture’s first display-free smart glasses have bone-conduction microphones to privately record your daily musings.
Inside two years of fraught AI data center debates in Pennsylvania.
加拿大一省政府就校园枪击案起诉OpenAI chinanews.com.cn
The British AI data center developer depends on tech giants Microsoft and Anthropic for most of its revenue.
This article is brought to you by CoolIT, an Ecolab Company . Beyond 250 kW a server rack can no longer be cooled by a hybrid approach of liquid and air. At this density a 70/30 liquid-air split leaves 75 kW of air load.
OpenAI研究员惊人言论：物理隔离无法阻止失控AI！ 智源社区
To build and deploy sophisticated robotics applications that can perceive, reason and act in dynamic environments, developers need new physical AI models and tools. The ROS open framework is a project from Open Robotics
GPT‑6 Astra allowed Parallel’s agents to research and synthesize labor-market data in half the time and at half the cost vs. prior models.
梦想是做能帮妈妈订机票的AI浏览器。
Hugging Face CEO：「太棒了」
With MiMo-V2.6-Pro, Xiaomi moves to the top of the openly available AI models and drastically undercuts the competition on price. What makes that possible is massive reinforcement learning that cost $2.62 million. But th
日前，智能经济30人论坛AGI测评中心发布最新一期《AGI市场观察》（2026年9月12日-18日）。报告以OpenRouter词元消耗数据、各厂商官方披露价格及Arena基准测评为基础，对国产大语言模型做周度量化追踪，本期围绕用量、份额、集中度、价格与性价比等方面给出观察。 数据显示，深度求索于9月10日下调DeepSeek V4 Flash及V4.1 Flash定价，后者周用量随之增长483.62%，位次由第9升至第2；腾讯Hy4
OpenAI内部模型24天再解百余道数学难题，引入独立顾问组回应学界质疑 新浪财经
PLUS: How to set up ChatGPT to write in your voice
9月22日，2026杭州云栖大会上，阿里云CTO李飞飞系统性阐述了Agentic Cloud 战略布局，阿里云将Model、Harness、Context作为核心构建场景，推出全新的云上算力底座和Agent 构建治理平台AgentCore、Agent Sandbox、新一代高性能存储CPFS等一系列新品，让 AI 真正进入生产流程、走向规模化价值创造。 随着2026年Agent的爆发式增长，越来越多的智能体正跨过“问答型”的分水岭，进入
📌 One-Sentence Summary OpenAI 承诺支持独立评估，提供训练、评估和部署的深度访问，并提出了四个评估优先事项。 📝 Summary OpenAI 宣布其承诺支持独立评估，提供训练、评估和部署的深度访问。该公司提出了四个评估优先事项：独立评估安全案例、评估关键保障措施、评估能力评估和独立调查关键偏离事件。 💡 Main Points OpenAI 致力于支持独立评估，提供训练、评估和部署的深度访问。 OpenAI
OpenAI outlines priorities and principles for rigorous, secure, and independent third-party AI safety assessments of frontier models and safeguards.
📌 One-Sentence Summary 一篇论文表明，通过名为「问题开局」的方法改进首次检索步骤，可将深度研究智能体在 BrowseComp-Plus 上的准确率最高提升 7.4 个百分点，代价仅是增加少量工具调用。 📝 Summary 该推文介绍了一篇论文，指出首次检索步骤对深度研究智能体至关重要。论文提出的「问题开局」方法在智能体开始搜索前运行：将问题拆解为线索，把每条线索转化为互补的搜索，汇总并重排结果，再用这一排序结果作为