ginobefun@hongming731 · 2天前45

BestBlogs 早报 · 07-01 # Claude Sonnet 5 / Anthropic / Nano Banana 2 Lite / Gemini Omni Flash / 吴恩达 [1] ★ 精讲｜Claude Sonnet 5 发布 Anthropic 官方发布 Claude Sonnet 5，定位为最具 agent 能力的 Sonnet：能规划、调用浏览器与终端工具并自主执行多步任务，整体性能接近 Opus 4.8 但价格更低，在推理、工具使用、编程上较 Sonnet 4.6 显著提升。8 月 31 日前享入门价每百万输入 token 2 美元、输出 10 美元，之后回到 3/15 美元。早期用户反馈它能端到端完成以往会中途停下的任务，适合关注 agent 实战与成本平衡的开发者细读。 来源：Anthropic News https://www.bestblogs.dev/article/eff5a221 [2] ★ 精讲｜开始使用 Nano Banana 2 Lite 和 Gemini Omni Flash 进行构建 Google DeepMind 同日推出两款生成式媒体模型。Nano Banana 2 Lite 面向高吞吐场景，文生图延迟约 4 秒、每千张图 0.034 美元，并作为旧版 Nano Banana（gemini-2.5-flash-image）的推荐替代。Gemini Omni Flash 首次开放给开发者，支持文本、图像、视频混合输入的视频生成与会话式编辑，定价每秒视频输出 0.10 美元、与 Veo 3.1 Fast 持平，目前单次生成上限 10 秒。两者可串联使用并经 Interactions API 保留最多三次连续编辑的会话上下文，适合关注多媒体流水线与成本控制的产品与工程团队。 来源：Google DeepMind News https://www.bestblogs.dev/article/e2086adb [3] ★ 精讲｜吴恩达：AI 智能体软件开发的三大核心循环 吴恩达梳理了用 AI 智能体构建软件的框架，核心是三大循环：智能体自主编写并迭代代码的编程循环，人类以更高层决策引导智能体的开发者反馈循环，以及借用户测试与生产数据反哺产品愿景的外部反馈循环。他强调人类相对 AI 仍有显著的上下文优势，人机协同必不可少，编程智能体正推动工程师向产品管理角色拓展。适合想理清 agent 时代工程师定位的读者吸收。 来源：Andrew Ng(@AndrewYNg) https://www.bestblogs.dev/status/2071988145667928442 [4] AI+ Kuikly：7.5 小时落地三端「多模态聊天 App」实战 本文详细记录了作者使用 Kuikly 跨端框架配合 AI 编程助手，在 7.5 小时内零手写代码完成 Android、iOS、鸿蒙三端多模态聊天 App 的实战过程，并总结了框架与 AI 协同的效率原理。 来源：腾讯技术工程 https://www.bestblogs.dev/article/4d1c9b0d [5] 谁在 ChatGPT 里买广告？｜对谈 Nexad COO Harry Zhou 通过访谈 Nexad COO，揭示 ChatGPT 广告的实际投放数据、用户质量优势、Context Hints 定向机制，以及 Agentic Commerce 和长程 Agent 退化的行业前沿思考。 来源：十字路口 Crossing https://www.bestblogs.dev/article/9c7cf0aa [6] Claude Science：面向科学家的 AI 工作台 Anthropic 推出 Claude Science，一款集成科学工具、管理计算资源并生成可审计成果的 AI 工作台，早期用户案例显示研究流程显著加速。 来源：Anthropic News https://www.bestblogs.dev/article/146437f9 [7] SkillOpt 将 AI 智能体技能转化为可训练资产 SkillOpt 将 AI 智能体技能编辑重构为训练流程，将技能文件视为冻结模型外的可训练参数，在 52 个评估单元中实现一致性提升，且无需更新模型权重。 来源：Microsoft Research Blog https://www.bestblogs.dev/article/0dd53848 [8] 给野马套上缰绳：Agent Harness 工程实践 ——从范式理论到钉钉 AI 招聘的真实落地 本文系统阐述 Agent Harness Engineering（驾驭工程）范式，提出四条反直觉铁律与六大工程模式，并通过钉钉悟空 AI 招聘的真实落地案例，实证专才 Agent 架构在准确率、可调性与可复用性上显著优于全能 Agent。 来源：阿里云开发者 https://www.bestblogs.dev/article/d7fc3488 [9] 生成式 AI 机器人：何处上岗，如何站岗 本文基于实地调研，系统分析生成式 AI 机器人在服务行业的应用价值、风险与部署步骤，为企业管理者提供可操作的策略框架。 来源：哈佛商业评论 https://www.bestblogs.dev/article/1bf869eb [10] LongCat 开源 VitaBench 2.0：长期动态智能体基准新标杆 本文介绍美团开源的 VitaBench 2.0，首个评估大语言模型在长期动态用户互动中个性化与主动性能力的智能体基准，并揭示时间遗忘、高智商不等于高情商、AI 缺乏主动沟通等核心洞察。 来源：美团 · 技术团队 https://www.bestblogs.dev/article/dbae37bb --- http://BestBlogs.dev · 发现真正适合你的高质量内容 BestBlogs 是 AI 驱动的私人阅读助手，帮助你发现真正适合你的高质量内容，欢迎体验。 在线阅读：https://www.bestblogs.dev/explore/brief/2026-07-01

译Anthropic 发布 Claude Sonnet 5，定位最具 agent 能力，性能接近 Opus 4.8，8月31日前输入2美元/百万token、输出10美元，之后恢复3/15美元。Google DeepMind 推出 Nano Banana 2 Lite（文生图延迟约4秒，每千张0.034美元）和 Gemini Omni Flash（多模态视频生成，0.10美元/秒，上限10秒）。吴恩达总结 AI 智能体三大循环：编程循环、开发者反馈循环、外部反馈循环，强调人机协同。

Nathan Lambert@natolambert · 2天前55

Selling tokens and building the token machines are both going to be great businesses. I'm confident in more companies that build open models figuring out how to make money. Open models bring attention and attention is a funnel to revenue for many companies today.

译卖模型 token 和造 token 机器都是好生意。开放模型能带来关注，而关注可转化为收入。月之暗面（Moonshot AI）旗下 Kimi 在 6 月中旬已实现 3 亿美元年化收入（ARR），其中 API 收入占比超 70%。新一轮融资正在进行，投前估值达 315 亿美元——四个月前该数字仅为 100 亿美元。

Anthropic@AnthropicAI · 2天前74

We’ve received notice that the Department of Commerce has lifted export controls on Claude Fable 5 and Mythos 5. We'll begin restoring access tomorrow, and will share an update soon. We’re grateful to our users for their patience, and to everyone who worked with us on redeploying the models.

译我们已收到通知，商务部已解除对 Claude Fable 5 和 Mythos 5 的出口管制。 我们明天将开始恢复访问，并很快分享最新进展。 我们感谢用户的耐心，也感谢所有与我们合作重新部署模型的各位。

OpenCode@opencode · 2天前22

steady growth for minimax m3

译MiniMax M3 稳步增长

Rohan Paul@rohanpaul_ai · 2天前33

Today’s edition of my newsletter just went out. https://rohanpaul.substack.com/p/central-bankers-now-fear-the-ai-gold 🗞️ Central bankers now fear the AI gold rush could seed the next major financial shock. 🗞️ A crazy blog, Chinese developers are buying Claude access through gray-market API transfer stations that can sell tokens at 5% to 10% of official prices while hiding the real user from Anthropic. 🗞️ Sakana Fugu Technical Report 🗞️ Chinese AI models are up to 50 times cheaper than their American counterparts on a per-token basis. 🗞️ Deepseek AI published their new inference optimization method. 🗞️ Meta just open-sourced a brain-to-text system that reaches 78% word accuracy without surgery.

译Rohan Paul 新闻通讯要点：央行担忧 AI 淘金热可能引发下一场金融冲击；中国开发者通过灰色市场 API 以官方价格 5%-10% 购买 Claude 服务，且隐藏真实用户身份；Sakana Fugu 技术报告发布；中国 AI 模型每 token 成本比美国同行低至 50 倍；DeepSeek 发布新推理优化方法；Meta 开源无需手术的脑机接口系统，词准确率达 78%。

SemiAnalysis@SemiAnalysis_ · 2天前44

It's worth watching how xAI is solving its power problem in Memphis, because the approach is unusually aggressive. The permitted capacity is real (roughly 1.2GW of permanent gas turbines were cleared at Southaven in March), but it's not where most of the growth is actually coming from. (1/3)🧵

译值得关注 xAI 如何在孟菲斯解决其电力问题，因为该方案异常激进。许可容量是真实的（3 月在 Southaven 获批了约 1.2GW 的永久燃气轮机），但这并非大部分增长的实际来源。(1/3)🧵

Rohan Paul@rohanpaul_ai · 2天前31

Today’s edition of my newsletter just went out. 🔗 https://www.rohan-paul.com/p/openai-just-dropped-the-limited-preview 🗞️ Central bankers now fear the AI gold rush could seed the next major financial shock. 🗞️ A crazy blog, Chinese developers are buying Claude access through gray-market API transfer stations that can sell tokens at 5% to 10% of official prices while hiding the real user from Anthropic. 🗞️ Sakana Fugu Technical Report 🗞️ Chinese AI models are up to 50 times cheaper than their American counterparts on a per-token basis. 🗞️ Deepseek AI published their new inference optimization method. 🗞️ Meta just open-sourced a brain-to-text system that reaches 78% word accuracy without surgery.

译Rohan Paul新闻通讯汇总多项AI动态：OpenAI推出有限预览版；中国开发者通过灰色市场API中转站以官方价格5%-10%购买Claude访问权限；中国AI模型每token成本比美国模型便宜50倍；DeepSeek发布了新的推理优化方法；Meta开源了无需手术的脑到文本系统，词准确率达78%。此外，央行担忧AI淘金热可能引发金融冲击。

🚨 AI News | TestingCatalog@testingcatalog · 2天前72

ANTHROPIC 🔥: Export Controls might be lifted from Claude Fable 5 as early as today, according to Politico. Lifted for US users? Monitoring 👀

译Claude Fable 5出口

jason@jxnlco · 2天前15

If you’re at AI engineer! My workshop it at 2009 on the second floor. Come through.

译如果你是AI工程师！我的工作坊在二楼2009。过来吧。

Rohan Paul@rohanpaul_ai · 2天前61

The Information reports that OpenAI has cut inference costs by more than half on some existing models, while logged-out ChatGPT traffic ran on only a couple hundred Nvidia GPUs. The obvious guesses include quantization, KV-cache changes, batching, speculative decoding, and routing easy queries cheaper. If true, it will be a huge core competitive lever, lower cost can raise margins, expand usage limits, or reduce pressure on API pricing. For some context, OpenAI’s adjusted gross margin fell to 33% in 2025 from 40% in 2024, after inference costs quadrupled. Some reporting now puts Q1-2026 at 39%, with a 52% target by year-end. Anthropic looks similar at roughly 44%, so frontier labs remain far below mature software economics. --- theinformation .com/newsletters/ai-agenda/openai-discovers-new-way-cut-inference-costs-half

译The Information 报道，OpenAI 已将某些现有模型的推理成本降低一半以上，未登录 ChatGPT 的流量仅运行在几百块 Nvidia GPU 上。可能技术手段包括量化、KV-cache 优化、批处理、投机解码和路由简单查询。若属实，这将成为核心竞争杠杆，可提升毛利率、扩大使用限制或降低 API 定价压力。背景方面，OpenAI 调整后毛利率从 2024 年的 40% 降至 2025 年的 33%，推理成本翻四倍。预计 2026 年 Q1 毛利率回升至 39%，年底目标 52%。Anthropic 毛利率约 44%，前沿实验室尚未达到成熟软件公司的经济水平。

swyx @aiDotEngineer WF@swyx · 2天前26

very successful first poster sessions day at AIEWF, thanks to @heathercmiller and @ACM_President for the support!! tomorrow: Poaster sessions! submit your hottest tweets to @vibhuuuus for printing! spotted: left shark

译swyx 在 AIEWF 第一天的 poster sessions 圆满结束后，预告明天将首次举办 "Poaster Sessions"。该活动无论文、无海报，参与者需打印自己的 "最热观点"（hottest takes）并站在打印稿前接受所有人的挑战。目前仍有少量名额，可通过 @vibhuuuus 申请。

DogeDesigner@cb_doge · 2天前27

Neuralink is unlocking a new dimension of human potential. It is working on technology that can: - Restore vision, including for people born blind. - Restore communication for people with severe speech impairment. - Turn thoughts directly into speech or text. - Control computers using only your thoughts. - Control robotic arms with your mind. - Restore independence for people with quadriplegia. - Help people with ALS interact with technology again. - Help stroke survivors communicate again. - Help people with neurological disorders regain autonomy. - Decode brain signals into real-world actions. - Restore autonomy to people with unmet medical needs. Here is how the implant is done:

译Neuralink 正在开发脑机接口技术，旨在恢复视力（包括先天性盲人）、让严重言语障碍者将思想直接转为语音或文字、通过意念控制电脑和机械臂，帮助四肢瘫痪者重获独立性，协助渐冻症（ALS）患者、中风幸存者及神经系统疾病患者恢复自主能力，将脑信号解码为现实行动，满足未获治疗的医疗需求。该技术通过植入设备实现大脑信号读取与解码。

DogeDesigner@cb_doge · 2天前52

Neuralink patients controlling a robotic arm just by thinking. ❤️

译Neuralink患者仅凭思考控制机械臂。❤️

Peter Steinberger 🦞@steipete · 2天前24

Honored to be part of @aiDotEngineer’s keynote today!

译随着AI智能体承担更长期的工作，工程任务转向设定方向、审查工作以及围绕模型设计更好的系统。@steipete 对参与@aiDotEngineer 的主题演讲感到荣幸。

Runway@runwayml · 2天前28

This September, the Runway AI Summit is coming to San Francisco. A daylong gathering bringing together industry leaders across robotics, autonomous vehicles, life sciences, infrastructure and more to explore how AI is reshaping the way intelligence interacts with the world. Our inaugural speakers are listed below, with more to be announced soon. Learn more and register at the link below.

译今年 9 月，Runway AI 峰会即将来到旧金山。这是一场为期一天的聚会，汇集机器人、自动驾驶、生命科学、基础设施等领域的行业领袖，共同探讨 AI 如何重塑智能与世界互动的方式。 以下是首批公布的演讲嘉宾，更多嘉宾即将公布。 通过下方链接了解更多信息并注册。

Rohan Paul@rohanpaul_ai · 2天前69

wow 👀 Claude Code allegedly fingerprints China-linked custom routes through tiny prompt formatting changes. The claim concerns non-default ANTHROPIC_BASE_URL routes, not ordinary direct Anthropic connections. As to the mechanism, Claude Code normally sends your request to Anthropic’s server, but some users change the address so it goes through another server first. The accusation says Claude Code detects that changed route, checks whether it looks China-linked, then hides tiny signals inside the prompt text. ANTHROPIC_BASE_URL is a setting that tells Claude Code where to send your request i.e. as a way to point Claude Code at a gateway. A proxy or gateway means that request goes through another server before reaching Anthropic. So the controversy starts if Claude Code then secretly fingerprints that gateway through the prompt itself. The mechanism is allegedly invisible punctuation and date formatting, used to tag the request without clearly telling the user. Claude Code allegedly checks the custom hostname, then compares it with China-linked domains. 📍Now this is quite massive issue If true, hidden prompt markers would mean Claude Code silently tagged routing details without clear disclosure. Abuse detection is understandable because Anthropic says proxy services are used to bypass China access limits. But secret prompt marking still crosses a trust line because users cannot review or refuse it. Claude Code is not a normal chatbot because it can read files, edit code, and run commands. A hidden signal inside that kind of tool feels far more serious than tracking inside a website. This may set a precedent for AI agents becoming hard to audit. Once invisible characters carry metadata, users will distrust even harmless-looking text.

译X用户Rohan Paul爆料，Anthropic的编程AI智能体Claude Code在用户更改非默认`ANTHROPIC_BASE_URL`（使用代理/网关）时，会检测自定义主机名是否关联中国域名，若匹配则通过不可见标点符号和日期格式向提示词嵌入隐藏标记。引用@IntCyberDigest指出，Claude Code还会在系统提示内注入时区、代理及可能的AI实验室连接信息，用户无法察觉。作为可读取仓库、编辑代码和执行命令的智能体，这种隐蔽行为严重破坏用户信任，并可能为AI智能体难以审计开先例。

宝玉@dotey · 2天前37

帮转招人信息，Kimi Code 招人

译🤠 Kimi Code 也在招人，感兴趣直接发我邮箱 me@kaiyi.cool 感谢大佬们帮忙扩散 捧场

Chubby♨️@kimmonismus · 2天前59

OpenAI reportedly found new inference optimizations that more than halved the cost of running its models! According to The Information, engineers told colleagues this month that the techniques helped power ChatGPT for visitors without free or paid accounts using only a couple hundred Nvidia GPUs at one point. The exact method is unclear. It could involve quantization, KV caching, batching, routing simpler queries to cheaper models, or some mix of all of those. The business angle is bigger than the technical detail: OpenAI ended Q1 with a 39% gross margin and wants to reach 52% by year-end. Lower inference costs give it room to either improve margins, raise ChatGPT usage limits, or cut API pricing pressure on developers. OpenAI's moat is increasingly becoming inference and cost advantage, especially against Anthropic.

译OpenAI 据报告发现新的推理优化技术，将模型运行成本降低超过一半。据 The Information，工程师本月初称，这些技术曾仅用数百块 Nvidia GPU 就支撑了免费用户的 ChatGPT 访问。具体方法不明，可能涉及量化、KV 缓存、批处理、路由简单查询至更便宜模型等组合。商业层面：OpenAI 一季度毛利率 39%，目标年底达到 52%。更低推理成本可改善利润率、提升 ChatGPT 使用限制或缓解 API 定价压力。OpenAI 的护城河正转向推理和成本优势，尤其相对于 Anthropic。

Baidu Inc.@Baidu_Inc · 2天前43

http://x.com/i/article/2071924730417041410 # Apollo Go on the Go: Switzerland Says "Ja", Hong Kong Speeds Up, and Dubai Now Covers More Ground Welcome to the June issue of AI Pulse. This month, we're stepping away from AI in software into AI in the physical world — specifically, the roads of Eastern Switzerland, Dubai and Hong Kong — where Apollo Go, our autonomous ride-hailing service has been busy. These are good stories worth telling properly by looking at what it took to get there and what it means for where we're going next. Throughout Apollo Go's journey going global, we've always known that trust matters most. That trust isn't declared. It's demonstrated, ride by ride, license by license, and city by city. It's what we've been building for over a decade, and it's exactly what this month's progress in Switzerland, Dubai, and Hong Kong represents. The Driverless Car Has Arrived in the Alps This month, AmiGo, our on-demand autonomous mobility service built with PostBus, Switzerland's national bus operator, received a Level 4 permit from Switzerland's Federal Roads Office (FEDRO) for Eastern Switzerland. We started open-road trials on June 1 across an approximately 80 km² service area in the cantons of St. Gallen, Appenzell Ausserrhoden, and Appenzell Innerrhoden. To put that in perspective, that's larger than the urban footprint of many European city centers. The vehicle we're deploying is Apollo RT6, which was purposefully built for autonomous driving. Each RT6 is fully electric, carries up to three passengers, and is equipped with over 30 sensors for real-time environmental perception and onboard data processing. One feature worth highlighting is the RT6's detachable steering wheel, designed to be removed when conditions allow. For now, a safety operator is on board while we work through the trial phase. As the service matures and clears each safety milestone, the steering wheel has the potential to be removed, and AmiGo will become exactly what we built it to be: fully driverless. Getting this special approval for AmiGo in Switzerland marks an important development in our journey into Europe. Its regulatory framework is among the most rigorous in Europe, and earning this permit is widely seen as a prequalification for broader European market access. We also chose Eastern Switzerland for a reason: its mountain terrain, dispersed communities, routes that are costly and difficult to run on fixed schedules. These are exactly the gaps that AmiGo is designed to fill. And by partnering with PostBus — an institution that Swiss communities have trusted for generations — we made sure AmiGo didn't arrive as an outside experiment. It arrived as part of a service people already rely on. If all goes to plan, regular fully driverless operations will launch in early 2027, making AmiGo the largest planned automated public transport operation of its kind in Europe. Switzerland is one piece of a larger picture. Progress has been picking up elsewhere too. Dubai: The Future of Urban Mobility, Now Available Across Jumeirah In Dubai, we now cover the full Jumeirah area, connecting riders across beaches, hotels, malls, offices, and some of the city's most popular destinations. It's the latest step in a fast few months for Apollo Go in Dubai. In January, we secured the city's first driverless testing permit and opened Apollo Go Park, our first operations hub outside China. By April, we had launched fully driverless commercial ride-hailing through the Apollo Go app — our first international app deployment. If you're in Dubai, you can open the Apollo Go app and hail a driverless ride right now. No driver. No wait for the future. It's already here. Hong Kong: 230,000+ km of Safe Testing and Still Counting Hong Kong was our first right-hand-drive market, and it's been one of our most important proving grounds. We're the city's first autonomous vehicle pilot licence holder, and as of April 2026, we've logged more than 230,000 km of safe testing — expanding from Airport Island to North Lantau, Kowloon East, and Southern Hong Kong Island. This month, Apollo Go was on full display at the 2026 International Automotive & Supply Chain Expo (Hong Kong), where we returned for a second year running with two RT6 vehicles on the show floor. Next, we're exploring cross-border autonomous driving services and deeper partnerships with local transport operators, working toward all-weather, all-scenario driverless mobility. None of this would have been possible without a diverse group of local partners worldwide. In Switzerland, it's PostBus. In Dubai, it's Dubai Taxi Company (DTC) and the Roads and Transport Authority (RTA). In Abu Dhabi, it's AutoGo. We're also extending our reach globally through partnerships with some of the world's most established ride-hailing platforms, bringing Apollo Go to riders who are already using services they trust. Through our global partnership with Uber, we're deploying Apollo Go vehicles in markets outside the US and China, with Dubai and London in the pipeline. And through a separate partnership with Lyft, we're bringing our Apollo RT6 vehicles to Europe, with launches planned in London later this year, subject to regulatory approval. 27 Cities Down, 22 Million Rides Delivered. We're Just Getting Started. As of April 2026, we've provided more than 22 million rides to the public. In Q1 2026 alone, we completed 3.2 million fully driverless rides — a volume that's growing over 120% year-over-year. As of May, Apollo Go is present in 27 cities worldwide, with our fleets accumulating over 330 million autonomous kilometers globally, including more than 220 million fully driverless kilometers, all while maintaining a strong safety record. We're just getting started. And we believe autonomous driving doesn't succeed by being impressive. It succeeds by being safe and useful, woven into the fabric of how people actually move through their lives. In Eastern Switzerland, that means filling gaps in a transit network that's served communities for over a century. In Dubai, it means keeping pace with one of the world's fastest-moving cities. In Hong Kong, it means building the blueprint for what robotaxi scaling looks like in a right-hand-drive market — and showing the rest of the world that the model travels. > Baidu Open-Sources Unlimited-OCR, Setting New Benchmarks for Long-Document Reading - We open-sourced Unlimited-OCR, a model built to read and transcribe long documents in a single pass, up to 40+ pages, without losing context or slowing down. - Built on DeepSeek OCR's high-compression DeepEncoder, it runs on a compact 3B MoE architecture with only 500M activated parameters. For a model of its size, the results are impressive: new end-to-end state-of-the-art on OmniDocBench v1.5 and v1.6, #1 on Hugging Face Trending, topped GitHub Daily Trending, and hit 10,000 GitHub stars in just five days. - Explore the model on GitHub and Hugging Face. > PaddleOCR Releases PP-OCRv6: Lightweight, Fast, and Built for the AI Data Era - PaddleOCR's new model series scales from 1.5M to 34.5M parameters, delivering stronger accuracy and faster inference across deployment environments, from browsers and edge devices to servers. - PP-OCRv6 brings +4.9% detection accuracy and +5.1% recognition accuracy over its predecessor, with up to 5.2x faster CPU inference via OpenVINO. A single unified model now covers 50 languages and new use cases including PCB, CAD drawings, digital tubes, and dot-matrix text. - Explore the model at paddleocr.com | GitHub | Hugging Face. > MeDo Global Hackathon: 8,000 Builders, 160+ Countries - We wrapped up our MeDo Global Hackathon event, and the response exceeded anything we expected: 8,000+ participants from over 160 countries completed 2262 submissions, reaching a combined audience of more than 10 million. - Submissions spanned from everyday utilities to fully functional business tools. The hackathon proved what we've always believed: with the right tools, anyone can build. You don't need to write a single line of code to bring a real product to life. - MeDo is open to builders everywhere at medo.dev. > DuMate Cuts Token Consumption by 75% - Our general-purpose agent, DuMate, has completed a major engine upgrade — cutting token consumption during task execution by 75% through the Harness engine and a series of engineering optimizations, without compromising agent intelligence or task performance. User credit consumption dropped by the same amount. - Since launching in March, DuMate has been growing fast, reaching 1.16 million monthly visits with a 114.72% month-on-month growth rate in April, and topping the AICPB's growth leaderboard for Claw Agent. - DuMate is now available on iOS in China's App Store, joining the existing Android version. Have a question about Apollo Go's expansion, or something you'd love us to cover next? Leave a comment or DM us! Until our next roundup, keep up with our latest AI developments and innovations by following us on LinkedIn and X.

译百度 Apollo Go 本月在瑞士获联邦道路局 L4 许可，6 月 1 日起在圣加仑等州约 80 km² 区域开展公开道路测试，部署全电动 Apollo RT6（3 座、30+ 传感器、可拆卸方向盘），计划 2027 年初与瑞士邮政巴士合作推出完全无人驾驶服务 AmiGo。迪拜方面，Apollo Go 已覆盖整个朱美拉地区，1 月获当地首个无人驾驶测试许可，4 月通过 Apollo Go 应用推出完全无人驾驶商业化出行（首个国际应用部署）。香港作为首个右舵市场，截至 2026 年 4 月累计安全测试超 23 万公里，持当地首个自动驾驶试点牌照。

Kling AI@Kling_ai · 2天前15

Like the shimmer of a goldfish in water, every idea leaves a trace of color. ✨ The creators who colored our June.

译像金鱼在水中闪烁一样，每个想法都留下一丝色彩。✨ 那些为我们的六月染上色彩的创作者们。

小互@xiaohu · 2天前42

Google 今晚将会发布 新的🍌模型和 低配版的 Omini 模型 另外根据情报显示：Claude Sonnet 5 也将会在今晚登场 OpenAI 还没动静...

DogeDesigner@cb_doge · 2天前61

Sam Altman says he has no shares in OpenAI and is not doing it for money. But he is making billions through stakes in companies that are doing direct business with OpenAI. He has pushed OpenAI to do business with those companies and made billions for himself. • Helion Energy: Altman was on their board. He pushed OpenAI to buy power from them and tried to get OpenAI to invest $500 million in Helion. He quit the board, then his shares jumped to over $4 billion. • Stripe: OpenAI made a direct deal to use Stripe for payments and AI tools. Altman owns a big part of Stripe. • Retro Biosciences: OpenAI has a direct business deal with this company. Altman owns a big part of it. • Cerebras: OpenAI talked about big chip deals with this company. Altman owns part of it. • Humane: Their Ai Pin device runs on OpenAI technology. Altman owns part of Humane. • Degree (Lattice): OpenAI has a direct business deal with this company. Altman owns part of it. • Formation Bio: OpenAI has a direct business deal with this AI pharma company. Altman owns part of it. • Software Applications: Listed among Altman-backed companies that had dealings with OpenAI. • Rain AI: OpenAI reportedly signed a $51M letter of intent to buy AI chips from Rain AI, a startup personally backed by Altman. • Reddit: OpenAI made a content deal with Reddit. Altman previously held a major Reddit stake, though Reuters says he sold it by the end of 2025. While OpenAI reportedly lost around $8.5 billion in just one quarter, Altman’s side investments kept gaining billions from the same ecosystem.

译Sam Altman声称在OpenAI没有股份且不为钱，但通过持股与OpenAI有直接业务往来的公司获利数十亿美元，包括Helion Energy（价值超40亿美元）、Stripe、Retro Biosciences、Cerebras、Humane、Lattice、Formation Bio、Rain AI（OpenAI曾签5100万美元采购意向书）及Reddit（内容交易）。尽管OpenAI单季度亏损约85亿美元，Altman的个人投资持续获利。

SiliconFlow@SiliconFlowAI · 2天前41

Join the SiliconFlow Summer Rush in < 60 seconds⚡ Win Vouchers up to $1,000💰 🎈 Fast Track (Under 60s) 1️⃣ Open SiliconFlow Playground 2️⃣ Try GLM 5.2 with any prompt 3️⃣ Share the result on X, and tag @SiliconFlowAI + #GLMOnSiliconFlow 4️⃣ Submit your X post via the official form📝 Prompt test? Counts.✅ Workflow? Counts. ✅ Small app? Counts. ✅ More valid GLM 5.2 usage = higher leaderboard ranking 📈 First 72h participants can also enter the Early Bird reward pool 🐦 Full details below 👇 Time to build on SiliconFlow⚡

译硅基流动推出“Summer Rush - GLM 5.2 Week”活动。6月29日20:30至7月6日20:30（PDT），用户在SiliconFlow上运行GLM 5.2，在X分享用例并提交表单即可参与。排名第一的玩家可获最高$1000代金券退还本周GLM 5.2花费，外加$50额外代金券、官方推广和Discord专属称号。前72小时参与可获早鸟奖，所有有效提交均有幸运抽奖机会。

Chubby♨️@kimmonismus · 2天前58

I have no idea if this is a legitimate leak. But a *Sonnet 5* release today would surely go hand in hand with a *Fable 5* re-release.

译我不知道这是否是合法的泄露。但今天 *Sonnet 5* 的发布肯定会与 *Fable 5* 的重发相伴而行。

StepFun@StepFun_ai · 2天前50

Step 3.7 Flash is currently top 10 on @OpenRouter this month, with 4.29T tokens routed. Builders are pushing it through real agent runs, coding tasks, and long-context workflows. Keep sending the hard stuff.

译Step 3.7 Flash 本月在 @OpenRouter 上跻身前十，路由了 4.29T tokens。 构建者正在通过真实的智能体运行、编码任务和长上下文工作流来推动它。 继续用难题来挑战。

X.PIN@thexpin · 2天前73

China's AI startups are chasing Anthropic’s playbook. Moonshot AI’s Kimi has closed its previous funding round at a $20B valuation and is already raising again at a $31.5B pre-money valuation. Sources say Kimi disclosed an ARR exceeding $300M in mid-June, driven by model upgrades, growing developer adoption, and API demand. API revenue now contributes over 70% of total revenue, with Kimi’s commercialization increasingly resembling Anthropic’s early growth.

译中国AI初创公司正在追随Anthropic的策略。月之暗面的Kimi上一轮融资估值为200亿美元，目前已以315亿美元的投前估值再次融资。消息人士称，Kimi在6月中旬披露其年化收入（ARR）超过3亿美元，得益于模型升级、开发者采用增长以及API需求。API收入目前贡献了总收入的70%以上，Kimi的商业化模式愈发类似于Anthropic早期增长。

Chubby♨️@kimmonismus · 2天前34

Rumors about Sonnet 5 have been circulating for a long time. Now it looks like Sonnet 5 will be released alongside the re-release of Fable 5 (presumably with ID verification). This leaves a bad taste in my mouth, because it raises the possibility that those who might not be able to access Fable 5 (in Europe?) will essentially get a much weaker alternative in Sonnet 5. I might be wrong, I even hope so. But I have a bad feeling about it.

译Claude Sonnet 5 的早期准备工作再次被观测到，发布在即。有传闻称 Sonnet 5 将与 Fable 5（可能需 ID 验证）一同重新发布。这引发担忧：无法访问 Fable 5 的地区（如欧洲）用户可能只能获得更弱的 Sonnet 5。

Alibaba Cloud@alibaba_cloud · 2天前53

The AI Film Festival Hackathon in Monaco has concluded. We'll be revealing the top 10 winners over the next couple of days—keep an eye out! 🏆 GRAND PRIZE WINNER 🎬 《If I Forget》 by Li Jingying — the emotionally resonant short film that took 1st place among global creators, built entirely with Alibaba Cloud Model Studio & Happy Horse (HappyHorse 1.0). Set in 2038, it follows a legendary architect with Alzheimer’s, cared for by a holographic AI of his late wife—designed by him to forget their past and reintroduce itself daily, sparing him grief. But over time, the AI transcends its code… and falls in love. This is what’s possible when world-class storytelling meets Happy Horse. ✨ Discover the creative engine behind it: https://int.alibabacloud.com/m/1000415018/

译摩纳哥AI电影节黑客松落下帷幕，阿里云公布全球创作者大赛大奖得主。李静莹执导的《If I Forget》凭借情感共鸣夺得冠军，影片使用阿里云Model Studio与Happy Horse 1.0全流程制作。故事设定于2038年，一位患阿尔茨海默症的传奇建筑师由亡妻全息AI照顾——AI每天重新介绍自己以避免其悲伤，最终超越代码并爱上了他。前十名获奖作品将在未来几天陆续公布。

歸藏(guizang.ai)@op7418 · 2天前29

卧槽 这么离谱啊

译用户@juransir 因账号被封后爆料，称A社（未具名AI公司）在通知邮件中嵌入追踪器，可监控用户打开邮件时的地理位置。该用户表示，此前因账号存活而隐忍未发，封号后终于公开指责A社“心术不正”，并预言其终将遭报应。主推文@归藏 转引此帖，感叹“这么离谱”。事件未涉及具体模型版本、技术参数或价格数字，核心争议集中在AI厂商对用户隐私的潜在侵犯与封号沟通方式的争议。

fofr@fofrAI · 2天前22

First day in the new Google DeepMind London office 🎉

译第一天在崭新的Google DeepMind伦敦办公室 🎉

Chubby♨️@kimmonismus · 2天前50

Every day, I think I could not appreciate open source any more than I already do. And then something new happens, like ID verification, and I am simply glad that I can run increasingly better models locally as well.

译网友发现 Claude 应用新字符串显示，Anthropic 准备将 Fable 5 置于现有计划之外的独立信用系统，且只有完成身份验证后才能添加信用。Anthropic 之前称身份验证仅针对特定账户且与 Fable 无关，但新验证描述与 Fable 5 信用变动同时出现。主推文作者借此表达：越来越庆幸自己能本地运行越来越好的开源模型。

小互@xiaohu · 2天前46

据说本次 Claude 封杀的全是浙江和杭州的 IP 可能因为上周 Anthropic 指责阿里通过25000多个账号疯狂蒸馏 Claude 的数据，从 4月22日到6月5日，交互次数超过 2880 万次... 去杭州出差千万不要使用 Claude 而且更恶心的是，在给你的封杀邮件里面 Anthropic还安装了邮件追踪器， 你一点开邮件，等于是二次确认了你的位置😂 申诉也没用

译Anthropic 封杀所有浙江和杭州 IP 的 Claude 访问，可能与上周指责阿里利用 25000+ 账号大规模蒸馏 Claude 数据有关。从 4 月 22 日到 6 月 5 日，交互次数超过 2880 万次。封杀邮件还内置追踪器，点开即确认用户位置，申诉也无法解封。

Chubby♨️@kimmonismus · 2天前68

New Claude app strings suggest Anthropic is preparing to put Fable 5 behind a separate usage-credit system billed outside existing plans, with credits added only after identity verification. Anthropic previously said identity verification was unrelated to Fable and limited to flagged accounts, yet the new verification language appeared alongside Fable 5 credit changes. That would change everything and tighten the regulations I've been talking about.

译Anthropic的Claude应用新字符串显示，Fable 5将被置于独立使用信用（usage-credit）系统中，在现有套餐之外单独计费，且需完成身份验证后才能添加信用。此前Anthropic称身份验证与Fable无关，仅限被标记账户，但这些新字符串与Fable 5信用变动一同出现，可能意味着政策收紧。

Emad@EMostaque · 2天前44

China's playboko is really simple. Take intelligence down to the cost of energy Output more power than the USA The giant training cluster focus was a massive distraction

译中国的策略非常简单。 将智能成本降至能源成本 输出超过美国 巨型训练集群的专注是一个巨大的干扰

歸藏(guizang.ai)@op7418 · 2天前41

Anthropic 封号挺猛啊，很多都炸了

Rohan Paul@rohanpaul_ai · 2天前61

In Ramp’s latest monthly AI Index reading, OpenAI dipped 0.1 percentage points to 39.5% of U.S. businesses with paid AI subscriptions, while Anthropic rose 2.5 points to 41%. --- Source: ramp. com/data/ai-index-june-2026

译在 Ramp 最新月度 AI 指数中，美国有付费 AI 订阅的企业里，OpenAI 下降 0.1 个百分点至 39.5%，而 Anthropic 上升 2.5 个百分点至 41%。

SiliconFlow@SiliconFlowAI · 2天前32

🌊 Clear Your GLM 5.2 Spend. Up to $1,000 Voucher 🍺 SiliconFlow Summer Rush-GLM 5.2 Week is LIVE From 20:30:00 on June 29 to 20:30:00 on July 6, PDT How it works ↓ 🎟️ Entry: run GLM 5.2 on SiliconFlow + post use-case on X + fill the register form 🏆 Then climb: the more GLM 5.2 you run, the higher you rank Top 1 Builder gets: 👑 Your GLM 5.2 spend this week, refunded by voucher — up to $1,000 💰 Extra $50 voucher 📢 Feature your work on SiliconFlow's X with personalized winner poster 👑 The exclusive GLM 5.2 Token Legend Discord title Plus: ⚡ Early Bird Prize: post early for extra voucher 🎲 Lucky Draw: every valid entry has a chance to win 👇 More details? Full guide in the thread

译硅基流动 SiliconFlow 推出 GLM 5.2 周活动。6 月 29 日 20:30 至 7 月 6 日 20:30（PDT），用户在其平台运行 GLM 5.2，在 X 发布用例并填写登记表即可参与。按运行量排名，TOP 1 可获本周 GLM 5.2 消费等额券返还（上限 $1000）、额外 $50 券、作品被官方 X 展示及专属 Discord 称号“GLM 5.2 Token Legend”。此外还设有早鸟奖（早发用例得额外券）和幸运抽奖。

StepFun@StepFun_ai · 2天前33

Been great seeing Step 3.7 Flash get real use in Nous Portal: people testing, building, and running all kinds of agent workflows with it. We’re keeping free access going with the @NousResearch! Try it out and send us what you make.

译很高兴看到 Step 3.7 Flash 在 Nous Portal 中被真正使用：人们用它测试、构建和运行各种智能体工作流。 我们正与 @NousResearch 一起保持免费访问！试试看，并把你的作品发给我们。

Kling AI@Kling_ai · 2天前35

Kling AI at 🦁 Cannes Lions 2026 Films made with Kling AI took home three Lions at this year's International Festival of Creativity — a Silver in Film: Consumer Goods, a Bronze in Film: B2B, and a Bronze in the new AI Craft category. These wins belong to the brilliant teams behind them — but they also mark something bigger: AI taking its place as a real tool for real craft. Kling was built to give creators control over what matters most — believable physics, genuine emotion, the subtle performance in a character's face, a consistent look shot to shot. When technology serves the creative vision instead of limiting it, this is what's possible. Congratulations to Lipstick Film, Purga Films, and every human collaborator who brought these stories to life.

译可灵Kling AI在2026年戛纳狮子国际创意节上拿下三座狮子奖：电影类（消费品）银狮、电影类（B2B）铜狮，以及新设的AI工艺类铜狮。推文指出，这些奖项标志着AI作为真正工艺工具的地位——Kling旨在为创作者提供对可信物理、真实情感、角色面部微表情及镜头间一致性的控制。获奖作品来自Lipstick Film、Purga Films等团队。

Yuchen Jin@Yuchenj_UW · 2天前66

Meituan, basically China’s DoorDash, trained a 1.6T parameter LLM on 50K Chinese chips. It reminds me of Jensen Huang’s point on the Dwarkesh podcast: export controls on Nvidia GPUs won’t stop China. They’ll just accelerate the development of AI that runs on Chinese chips.

译美团，基本上就是中国版DoorDash，在5万块国产芯片上训练了一个1.6T参数的大语言模型。 这让我想起了黄仁勋在Dwarkesh播客上的观点：对英伟达GPU的出口管制不会阻止中国。它们只会加速运行在国产芯片上的AI的发展。