全部 AI 动态

🚨 AI News | TestingCatalog@testingcatalog · 1小时前46

Morning Brief feature seems to be rolling out on Claude for some users. Morning Brief runs on scheduled tasks and guides users through customizing preferences and Connectors to pull data. Did you get it too? 👀

译Morning Brief 功能似乎正在向部分 Claude 用户逐步推出。 Morning Brief 基于定时任务运行，引导用户自定义偏好设置和 Connectors 以拉取数据。 你也收到了吗？👀

🚨 AI News | TestingCatalog@testingcatalog · 1小时前50

Claude got an updated unified memory across Claude Chat and Cowork. Also, a new toggle to keep memories about sensitive topics has been added to the Settings menu. Memoriesed topics are now split into separate categories. Memory now has a dedicated tab in Claude Settings. Users can now tell Claude what to remember directly from the memory browser UI.

译Claude 更新了跨 Claude Chat 和 Cowork 的统一记忆功能，用户可决定记忆内容。设置中新增敏感话题记忆开关，记忆按类别拆分，并在设置中新增独立标签页，用户可直接在记忆浏览器界面告诉 Claude 要记住什么。

Greg Brockman@gdb · 1小时前46

you can now have chatgpt work securely sign into accounts:

译ChatGPT Work 现可调用其计算机和浏览器，在网页及移动端安全登录网站，全程不接触用户名或密码。用户可让其代办理水电开通、预约 DMV 或护照、查询保险报销、预约医生、处理车辆登记续期、提交报销单据等 20 余项事务。

OpenRouter@OpenRouter · 2小时前42

New inference provider: @makora_ai is now live on OpenRouter. Starting with: • DeepSeek V4 Flash • GLM 5.2 FP8 • GLM 5.2 NVFP4 • Kimi K3 Route through Makora: https://openrouter.ai/provider/makora

译新的推理服务商：@makora_ai 现已上线 OpenRouter。 首批提供：• DeepSeek V4 Flash • GLM 5.2 FP8 • GLM 5.2 NVFP4 • Kimi K3 通过 Makora 路由：https://openrouter.ai/provider/makora

Greg Brockman@gdb · 3小时前46

by popular demand, $100 business seat now available in chatgpt! no 5-hour limit, more usage, etc..

译应大家要求，ChatGPT 现已推出 100 美元的商务席位！无 5 小时限制，更多用量等。

Boris Cherny@bcherny · 3小时前39

A small improvement: memory is now simpler and more powerful. Enjoy!

译一个小改进：记忆功能现在更简单、更强大。尽情享受吧！

Tibo@thsottiaux · 3小时前40

So much demand for this one. Works similar to the Pro $100 plan but designed for teams and small companies. ✅ All ChatGPT, ChatGPT Work, and Codex features ✅ Connect to Google Workspace, Slack, GitHub, Microsoft 365, and more ✅ Secure workspace with SAML, SSO, and MFA ✅ Centralized billing and administration ✅ Usage analytics and spend controls ✅ No 5h limits

译OpenAI 推出 ChatGPT Business Premium Seats，面向团队和小公司的 $100/席位套餐，功能对标 Pro $100 计划。

OpenAI Developers@OpenAIDevs · 4小时前37

The ChatGPT extension now supports more of the browsers you already use. Bring open tabs into a task so Codex can work from the context in front of you.

译OpenAI 宣布 ChatGPT 浏览器扩展现已支持 Microsoft Edge、Brave、Opera 和 Vivaldi 浏览器。用户可将浏览器标签页上下文带入 ChatGPT 桌面端任务，让 Codex 基于当前页面内容工作，并可通过侧边聊天使用完整 ChatGPT 功能。Opera 的侧边聊天功能即将推出。

Sam Altman@sama · 4小时前22

we made a chip and it is fast

译我们造了一颗芯片，而且它很快。

Sherwin Wu@sherwinwu · 4小时前36

The $100 ChatGPT Business seat is here now! I've lost count of how many startup friends of mine have been asking for this. Hoping this is a good middle ground, especially for folks at startups.

译OpenAI 推出 ChatGPT Business Premium Seats，每个席位月费 100 美元，面向小企业和初创公司。该计划提供更强大的工具、更快的工作流，以及以往大公司才能使用的功能，并可按团队需求灵活扩展。

OpenAI@OpenAI · 4小时前42

Introducing ChatGPT Business Premium Seats The new $100 Premium seat is a game changer for small businesses and startups—giving lean teams better tools, faster workflows, and capabilities once reserved for big companies. A flexible plan that scales with your team’s ambition. https://chatgpt.com/pricing/?type=team

译推出 ChatGPT Business Premium 席位 新的 100 美元 Premium 席位对小企业和初创公司来说是一个颠覆性的改变--让精简团队获得更好的工具、更快的流程，以及曾经只有大公司才能拥有的能力。 一个灵活的方案，随你团队的雄心一同扩展。 https://chatgpt.com/pricing/?type=team

OpenCode@opencode · 4小时前50

Grok 4.6 is now available on OpenCode Go you get one hundred 69 requests every 5 hours

译Grok 4.6 现已可在 OpenCode Go 上使用 每 5 小时可获得一百次 69 请求

🚨 AI News | TestingCatalog@testingcatalog · 5小时前39

Keenable has launched an independent web search API for AI labs and agents. It runs on its own crawler, index, and ranking stack across more than 100 billion documents. It has: 26M seed round led by Accel p95 latency under 250ms from US East, The company also launched Time Machine, that rewinds search results to how the web looked at a chosen point in time.

译Keenable 为 AI 实验室和智能体推出独立网页搜索 API，基于自研爬虫、索引和排序栈，覆盖超 1000 亿文档，获 Accel 领投 2600 万美元种子轮，美国东部 p95 延迟低于 250ms。同步发布 Time Machine 功能，可将搜索结果回退至指定时间点的网页状态。Web Search API 与 Web Query Language 现已上线，9 月底前免费。

Suno@suno · 5小时前54

Suno Studio 2.0 bridges the gap between traditional DAW expression, like MIDI input and real time audio effects, and our latest tech. Play it, tweak it, perfect it. 🎹🎧 #Suno #SunoStudio2 #MusicTech #DAW

译Suno Studio 2.0 弥合了传统 DAW 表现方式（如 MIDI 输入和实时音频效果）与我们最新技术之间的差距。 演奏它、调整它、完善它。🎹🎧 #Suno #SunoStudio2 #MusicTech #DAW

Perplexity@perplexity_ai · 5小时前48

New research: Portable Computer is a local-first agent for private and cost-effective work. With an on-device 27B model, our harness scores 82.6% on real knowledge work, beating open-source harnesses Pi and Hermes. Our post-trained PPLX 27B reaches 85.4%.

译新研究：Portable Computer 是一款本地优先的智能体，用于私密且经济高效的工作。 凭借设备端 27B 模型，我们的框架在真实知识工作中得分 82.6%，超越了开源框架 Pi 和 Hermes。我们后训练的 PPLX 27B 达到 85.4%。

Google Gemini@GeminiApp · 5小时前54

Dictate, summarize files, and rewrite copy directly into any window with your voice in the Gemini app for macOS.

译在 macOS 版 Gemini 应用中，直接用语音在任意窗口内听写、总结文件并改写文案。

Andrew Ng@AndrewYNg · 5小时前精选67

OpenWorker -- an open source agent that doesn't just chat but completes tasks on your laptop -- just released a new version with many features for security workflows. After our initial release, many users found it especially useful for cybersecurity. Attackers are already using AI; OpenWorker is committed to giving defenders the same leverage. Running an agent requires both (i) A model and (ii) A harness (the software around the model). Because the OpenWorker harness is fully open source, security teams can audit it to make sure we haven't built any backdoors that exfiltrate your code and data to some company or even a foreign adversary. OpenWorker now comes with built-in cybersecurity agents for (i) Scanning your code for vulnerabilities. (ii) Scanning dependencies for supply chain injections. (iii) Checking your cloud security configuration for attack surfaces. This enables developers to do much more security work before deployment (part of what's called the "shift left" movement). You choose the model: you can run open weight models fully locally so sensitive code never leaves your machine. This helps with legitimate security work (like reproducing a known exploit to defend against it) that can trigger refusals in leading closed models. Or use your ChatGPT subscription, or stealth preview models like Ox Alpha, or any model via API key. Thanks also to all the open source contributors! Join work with @rohitcprasad so please follow him too to get more frequent updates. Try it out: https://openworker.com/ Code: https://github.com/andrewyng/openworker

译Andrew Ng 旗下开源智能体 OpenWorker 发布新版，强化安全工作流。其 harness 完全开源，安全团队可审计无后门。新版内置代码漏洞扫描、依赖供应链注入检测和云安全配置检查三类网络安全智能体，并支持本地运行开源权重模型以保护敏感代码。

lauren@poteto · 5小时前43

just shipped deeplinks for Grok @Bot on desktop (mobile coming soon)! try installing the X plugin: grokbot://app/v1/plugin/add?id=49086599 and pstack of course: grokbot://app/v1/plugin/add?id=9717366

译刚刚为 Grok @Bot 在桌面端上线了 deeplinks（移动端即将推出）！ 试试安装 X 插件：grokbot：//app/v1/plugin/add？id=49086599 当然还有 pstack：grokbot：//app/v1/plugin/add？id=9717366

SemiAnalysis@SemiAnalysis_ · 5小时前37

OpenAI's new Jalapeño chip announced at Hot Chips beats Vera Rubin's July results on Output Throughput per MW. (1/7)🧵

译OpenAI 在 Hot Chips 大会上发布的全新 Jalapeño 芯片，在每兆瓦输出吞吐量上超越了 Vera Rubin 七月份的结果。（1/7）🧵

Chubby♨️@kimmonismus · 5小时前50

interesting detail in OpenAIs blogpost to "Jalapeño": GPT-Astra used Codex to write and optimize low-level kernels that brought three previously unplanned open-weight models to high performance on OpenAI’s Jalapeño chip within two months. For selected attention and mixture-of-experts blocks, its implementations ran 1.5–1.8× faster than existing human-expert-written code. looks like astra is a really really good model.

译OpenAI 博客透露，GPT-Astra 使用 Codex 编写并优化低层内核，让三款此前未规划的开放权重模型在两个月内于 Jalapeño 芯片上达到高性能。在选定的注意力与混合专家模块中，其实现比现有的人类专家代码快 1.5-1.8 倍。Jalapeño 在 OpenAI 自测中已超越 Nvidia GB200/GB300，计划 2026 年底部署。

Tibo@thsottiaux · 6小时前40

Excited about our Jalapeno results today. An incredible achievement from the team, taking a new chip from concept all the way to very impressive performance on real workloads in the lab. Tomorrow’s fast will feel like today’s ultrafast. As I’ve mentioned before, we’re pushing to bring this to as many people as possible. We’ve seen a TON of demand for /ultrafast, made possible by our deep partnership with Cerebras and their unique hardware, which will push the frontiers of tomorrow's ultrafast even further. I’m very much looking forward for the collaboration to continue pushing the absolute limits of how fast we can run our most capable models on Cerebras and bring this to the most demanding customers.

译OpenAI 首款自研推理芯片 Jalapeño 已在实验室完成实测，从概念到真实负载表现优异。该芯片在更高吞吐和更低延迟上实现"每瓦特更多智能"，并与 Cerebras 深度合作推动 /ultrafast 高速推理，满足大量客户需求。

Replit ⠕@Replit · 6小时前17

In Free Mode you can actually do more, while worrying about less. It’s really that simple.

译在免费模式下，你实际上可以做得更多，同时担忧得更少。就是这么简单。

OpenRouter@OpenRouter · 6小时前43

TIP 💡 Our :nitro and :floor model variants tap into service tiers, not just sorting Example: "~openai/gpt-latest:nitro" has many features packed into 1 model slug: 1. ~latest alias resolves to the latest gpt model 2. :nitro sorts endpoints by measured throughput 3. (NEW) :nitro admits priority service tier endpoints into the pool, competing on speed alongside the defaults

译TIP 💡 我们的 ：nitro 和 ：floor 模型变体接入服务层级，而不仅仅是排序 示例："~openai/gpt-latest：nitro" 将许多功能打包进一个模型标识中： 1. ~latest 别名解析为最新的 gpt 模型 2. ：nitro 按实测吞吐量对端点排序 3. （新功能）：nitro 将优先服务层端点纳入池中，与默认端点一同比拼速度

Krea@krea_ai · 6小时前58

wan 3.0 is now on Krea. 20 reference images and 30 second generations with audio at a fraction of the cost of any other models of this quality. try it now 👇

译Wan 3.0 现已登陆 Krea。 支持 20 张参考图像，以及 30 秒带音频的生成，成本仅为同类质量模型的零头。 立即体验 👇

Thariq@trq212 · 6小时前37

excited to share more on how we’re making Claude Code more hackable soon

译Anthropic 正提升 Claude Code 的可扩展性，将支持用户轻松使用 Agents.MD 或修改系统提示词。团队认为不同模型族的系统提示词对性能影响显著，且 Claude Code 已为各模型配置不同提示词。短期内用户可在 Claude.MD 中通过 @Agents.MD 调用。

OpenAI@OpenAI · 6小时前32

Since announcing Jalapeño, our first custom inference chip, we’ve been testing it and the system around it. The results show a major advance: more intelligence from every watt and faster responses, delivering both higher throughput and lower latency in one architecture without sacrificing efficiency.

译自发布我们的首款定制推理芯片 Jalapeño 以来，我们一直在对其及周边系统进行测试。 结果显示这是一项重大进步：每瓦特带来更多智能，响应速度更快，在同一架构中同时实现更高吞吐量和更低延迟，且不牺牲效率。

Claude@claudeai · 6小时前同事件53

Claude now has one memory across chat and Claude Cowork, and you decide what's in it. Hand Cowork a task and it starts from what Claude already knows from your chats: the project you talked through, your manager's preferences, or the client from last quarter.

译Claude 现在在聊天和 Claude Cowork 之间拥有统一的记忆，由你决定其中包含什么内容。 把任务交给 Cowork，它会从 Claude 在聊天中已经了解的信息开始：你讨论过的项目、你经理的偏好，或者上个季度的客户。

Peter Steinberger 🦞@steipete · 7小时前17

This week.

译本周。 【引用 @realcaptmarbles】：OpenClaw 看起来回来了，很期待。发布吧！

Aravind Srinivas@AravSrinivas · 8小时前20

On showing an early demo of Portable Computer on DGX Spark to Jensen, he was kind to gift us a DGX Station, a beast of a local computer that can serve even frontier models like GLM 5.3. Unmetered frontier intelligence running on your own local hardware coming soon!

译在DGX Spark上向Jensen展示Portable Computer早期演示时，他慷慨地送了我们一台DGX Station--一台强大的本地计算机，甚至可以运行GLM 5.3这样的前沿模型。无限制的前沿智能即将在你的本地硬件上运行！

🚨 AI News | TestingCatalog@testingcatalog · 7小时前39

OpenAI announced initial performance results of their own Jalapeño chip for AI inference. 1.5–1.9× more AI work per watt 1.7–3.6× lower end-to-end latency 2.1–4.1× higher performance on highly interactive workloads Looking forward to seeing this deployed!

译OpenAI 公布自研 AI 推理芯片 Jalapeño 的初步性能结果。在 InferenceX 上对三个公开模型的测试显示，其每瓦 AI 工作量提升 1.5-1.9 倍，端到端延迟降低 1.7-3.6 倍，高交互负载性能提升 2.1-4.1 倍。

🚨 AI News | TestingCatalog@testingcatalog · 7小时前35

Google announced its first plugins for Gemini Enterprise, tailored for legal and finance use cases. These plugins combine selected Apps and Skills under purpose based bundle instructions. It is also expected that bundles eventually will arrive to consumer Gemini. What worries me here is that we got only two. Soon? 👀

译Google 为 Gemini Enterprise 推出首批面向法律与金融场景的插件，将精选 Apps 与 Skills 按用途组合成捆绑包。未来捆绑包有望扩展至消费版 Gemini，但当前仅发布两个。合作生态涵盖 Accenture、Deloitte、KPMG 等全球系统集成商。

elvis@omarsar0 · 8小时前40

Big launch from @KeenableAI. The team built its own crawler, index, and query language for agents in under a year. I think owning the full search stack gives them room to improve quality, latency, and cost together as agent workloads grow. Worth checking out.

译Keenable AI 正式发布，打造 AI 原生的开放网络知识索引，让模型和智能体可查询、推理并持续学习实时网络。其 Web Search API 与 Web Query Language 已上线，9 月底前免费。公司获 Accel 和 Conviction 2600 万美元种子轮融资，团队来自 Yandex Search、Amazon AGI、X 和 Perplexity。

Lee Robinson@leerob · 8小时前38

We just increased the included usage of Grok models in Cursor! Demand has been increasing with the launch of Grok 4.6. We already doubled limits last month with more compute. And now we're raising limits again permanently. Enjoy!

译我们刚刚提高了 Cursor 中 Grok 模型的包含用量！ 随着 Grok 4.6 的发布，需求一直在增长。上个月我们已经通过增加算力将限额翻倍。现在我们再次永久提高限额。 尽情使用吧！

Aravind Srinivas@AravSrinivas · 8小时前37

In a compute and power-constrained world, a good chunk of agentic inference needs to move to local hardware. A drastic version of that is a fully local agent runtime, where the model (orchestrator and subagents) and the harness run locally. Portable Computer from Perplexity is this. Launching today for @nvidia DGX Spark.

译Perplexity 今日在 NVIDIA DGX Spark 上推出 Portable Computer，这是 Perplexity Computer 的完全本地版本，编排 LLM、子智能体 LLM 与智能体 harness 全部在本地硬件运行，无云端依赖。该产品面向算力与电力受限环境，将大量智能体推理迁移至本地设备。

elvis@omarsar0 · 8小时前34

Can't wait to have always-on agents running locally 24x7. Local agents are something Perplexity seems to be chasing these days aggressively, and it makes a ton of sense. There is a huge market for personalized AI agents.

译Perplexity 推出本地优先智能体栈 Portable Computer，运行于 NVIDIA DGX Spark，提供一键本地推理设置与优化的智能体体验。Elvis Saravia 看好本地智能体前景，认为始终在线、本地运行的个性化 AI 智能体存在巨大市场。

Greg Brockman@gdb · 8小时前41

inference numbers published for jalapeno, team did an amazing job https://openai.com/index/jalapeno-first-results/

译OpenAI 公布首款自研推理芯片 Jalapeño 的实测性能：在基于 GPT-OSS 120B 的公开基准 InferenceX 上，其每千瓦峰值吞吐量更高、token 延迟更低，并在 DeepSeek R1 和 Kimi K2 上表现强劲。AI 参与设计使其从初始设计到流片仅用九个月，计划年底前部署，Gen 2 和 Gen 3 已在开发中。

Perplexity@perplexity_ai · 9小时前53

Today we’re launching Portable Computer on @NVIDIA DGX Spark. Portable Computer is a fully local version of Perplexity Computer, where the entire runtime: orchestrator LLM, subagent LLM, agent harness all run on your local hardware. No cloud dependency.

译今天我们正式在 @NVIDIA DGX Spark 上推出 Portable Computer。 Portable Computer 是 Perplexity Computer 的完全本地版本，整个运行时--编排 LLM、子智能体 LLM、智能体框架--全部在你的本地硬件上运行，不依赖云端。

🚨 AI News | TestingCatalog@testingcatalog · 9小时前同事件58

APPLE 🔥: A new Mac Studio powered by M6 and M5 Ultra chips have been announced along with Mac mini, featuring M6 and M5 Pro. "M6, Apple’s first 2 nm chip, features a larger, more powerful 12-core CPU, 12-core GPU, and Dual 16-core Neural Engine, while M5 Ultra is Apple’s first quad-die architecture and its most powerful chip ever" "M5 Ultra, the ultimate powerhouse for pro and AI workloads, uses next-generation UltraFusion technology to form a quad-die architecture for the first time in an M-series system on a chip (SoC). The chip includes an up-to-36-core CPU and up-to-80-core GPU with a massive 1.2TB/s of unified memory bandwidth, 50 percent more than M3 Ultra." Mac mini with M6 delivers: Up to 13.5x faster LLM prompt processing in LM Studio when compared to Mac mini with M1, and up to 4.8x faster than M4. Mac mini with M5 Pro delivers: Up to 8.5x faster LLM prompt processing performance in LM Studio when compared to Mac mini with M2 Pro, and up to 4x faster than M4 Pro. I need a new Mac Studio cluster 👀

译苹果发布搭载M6和M5 Ultra芯片的新款Mac Studio及Mac mini。M6为苹果首款2nm芯片，配备12核CPU、12核GPU及双16核神经引擎；M5 Ultra采用首款四芯片封装架构，最高36核CPU、80核GPU，统一内存带宽达1.2TB/s。Mac mini（M6）在LM Studio中LLM提示词处理速度较M1快13.5倍，较M4快4.8倍。

🚨 AI News | TestingCatalog@testingcatalog · 8小时前42

PERPLEXITY 🔥: Perplexity Pro and Max subscribers can now use a new Portable Computer on DGX Spark from NVIDIA, powered purely by local models! Available models: PPLX 27B is a new post-trained model from Perplexity. Qwen 3.8 27B Nemotron 3.5 Lightning (coming soon) They really did it! 🤖

译PERPLEXITY 🔥：Perplexity Pro 和 Max 订阅用户现可在 NVIDIA DGX Spark 上使用全新的 Portable Computer，完全由本地模型驱动！ 可用模型：> PPLX 27B 是 Perplexity 新推出的后训练模型。> Qwen 3.8 27B > Nemotron 3.5 Lightning（即将推出） 他们真的做到了！🤖

elvis@omarsar0 · 9小时前35

Recommended read. Proactive agents are a whole new and underexplored category. This microharness is about having a persistent, always-on agent that's always thinking and doing important work. Proactive agents are exciting because they unlock new ways to work with agents and unlock productivity. My proactive harness has a similar functionality that stays running in the background, monitoring and tracking work across projects/tasks, and then, when it has a genuinely useful idea or suggestion, it activates. It learns my preferences as I interact with it over time, so it gets better at recommendations with every interaction.

译Headlong 是一个开源微框架，用于构建持续思考的主动型智能体，核心代码不足 10K 行 Bash。与反应式框架不同，Headlong 智能体永不休眠，通过自我引导循环持续生成想法，并自主决定何时回复。内部运行显示其后台思考成本约 $1-$2/小时，曾自主发现并修复一个 bug，全程 48 分钟无人工介入。