Live intelligence
The AI signal.
Minus the noise.
A chronological wire of AI developments selected for substance. Scan the reason, context and source without handing database access to the browser.
-
October 11
Sunday- Aravind Srinivas
推文涉及疑似新模型 OpenAI bel 的发布信息（传闻12月发布）并提及 fable 5.5，属于疑似新模型发布讨论且包含具体模型名。
openai bel will mog everyone google employee posted about an ai dinner in nyc, and consensus was that fable 5.5 will be stunning, but openai will beat it with bel in december for now anthropic is winning and fable 5.5 no doubt will be the best model too but looks like openai been cooking a beast…
bel fable 5.5 OpenAI Anthropic Open the source post in a new tab - Aravind Srinivas
Every Agent 产品正式发布/上线，包含具体产品名和发布动作
How the @Every Agent works: Tag @Every in any Slack channel, tell it what needs to get done, and it completes the task in the thread. It learns your company as it goes: When something isn’t right, you correct it in the thread, and it can save the correction for next time. Your Every Agent shows up…
Every Agent Every Open the source post in a new tab - Aravind Srinivas
Microsoft 365 Copilot 接入 Anthropic 旗舰模型 Claude Opus 5.5，属于具体模型接入/更新事件
所以聪明人都去闲鱼买 Microsoft 365 了？ 为了在 Word 里面用上 Claude 模型，也真是够拼的😄但这写不了代码啊，还不如中转站？ > [@MinLiBuilds](https://twitter.com/MinLiBuilds): > 🚨 全体起立！ 我发现Claude Opus 5.5，不用担心被封的野方法！ > > 居然藏在 Microsoft Word 里！ > > 我试了，真的可以。 > > 🔥 Microsoft 365 Copilot 在9月22日已经接入 Anthropic 旗舰模型 Claude Opus 5.5，直接在 Word…
Claude Opus 5.5 Microsoft 365 Copilot Anthropic Microsoft Open the source post in a new tab - Aravind Srinivas
OpenAI ChatGPT Pro 订阅套餐用量与定价策略重大调整（Pro 用量削减、新增 $500 Ultrafast 档、拟恢复 5 小时限制），属于产品重大变更事件
🚨 OpenAI is losing all its AURA. Again. And again. One questionable decision after another, and somehow, it keeps getting worse. → Pro 200 usage cut from 20x to 10x. → Same $200 price. Less usage. → A new $500 tier for more usage and Ultrafast. → All new Features like Ultrafast and ChatGPT Dots…
OpenAI Pro 200 Ultrafast ChatGPT Dots Open the source post in a new tab - Aravind Srinivas
Copilot Studio Plugin Registry Skills 进入 preview，属于 AI 产品重大功能更新
Microsoft is pulling the whole ecosystem together. Skills through the Plugin Registry are now in Copilot Studio preview. Build once. Reuse across agents and supported Copilot experiences. Govern centrally. This is how we win.…
Plugin Registry Copilot Studio Microsoft Open the source post in a new tab - Aravind Srinivas
包含疑似新模型 Opus 5.6 / Fable 5.5 的访问爆料，属于疑似新模型发布
hate when it gets stuck like this for minutes with nothing this is my 4th retry in a row trying to get output from next Opus or Fable 5.5 model got stuck anyone know a fix? > [@HarshithLucky3](https://twitter.com/HarshithLucky3): > Looks like I got access to next Opus model (Opus 5.6) or Fable 5.5 >…
Opus 5.6 Fable 5.5 Open the source post in a new tab - Aravind Srinivas
推文包含具体模型名 GLM 5.3 Flash 及其在 TensorFold 上新 EXL3 量化版本的发布信息
👇🏼Don’t take anyone’s word, try it yourself. > [@MiaAI_lab](https://twitter.com/MiaAI_lab): > It's not hype. It's not made up. > > GLM 5.3 Flash on TensorFold with the new EXL3 quants delivers better speed AND better quality. > > Try it yourself. > One recipe, works on 2 to 4 DGX Sparks. > > Link…
GLM 5.3 Flash TensorFold EXL3 Open the source post in a new tab - Aravind Srinivas
疑似Anthropic新模型Opus 5.5/5.6 checkpoint上线，属于疑似新模型发布
Damn, what is Anthropic cooking right now? Are requests already being routed to a new Opus 5.5 checkpoint, or even something like Opus 5.6? Mine seem to be hitting the new one, and it looks really, really good. > [@koltregaskes](https://twitter.com/koltregaskes): > Fable 5.5, is that you? 👀 > >…
Anthropic Opus 5.5 Opus 5.6 Fable 5.5 Open the source post in a new tab - Aravind Srinivas
疑似 Anthropic 发布新的 Opus 5.5 / Opus 5.6 模型版本，符合疑似新模型发布条件
Damn, what is Anthropic cooking right now? Are requests already being routed to a new Opus 5.5 checkpoint, or even something like Opus 5.6? Mine seem to be hitting the new one, and it looks really, really good. > [@koltregaskes](https://twitter.com/koltregaskes): > Fable 5.5, is that you? 👀 > >…
Opus 5.5 Opus 5.6 Anthropic Fable 5.5 Open the source post in a new tab - Aravind Srinivas
AI监控工具 sparkDash 发布 2.0 及后续多个版本更新，符合AI工具重大功能更新条件
sparkDash 是 DGX Spark 的开源监控面板，2.0 在 10/9 发布，2.31 在 10/10 发布，目前已经到 2.40。这两天的变化按版本列一下： 1. 2.0 整个界面重做了：侧边栏加命令面板（Ctrl/⌘ K），四套主题。Overview 每张卡片上直接是显存、GPU、温度、功耗和 tok/s 2. 2.0 可以在面板里启动和停止自己的模型，登记一个带 http://start.sh 和 http://stop.sh 的目录就行，启动输出在界面里实时滚动。另外新增 Benchmarks 板块，有 Decode、Prefill、Quality 和 Tool Eval…
sparkDash DGX Spark MiaAI-Lab Open the source post in a new tab - Aravind Srinivas
疑似 Anthropic 新模型 Opus 5.6 被通过 Opus 5.5 路由测试，包含具体模型名与疑似推出事件
Opus 5.5 Routing to Opus 5.6 - Some accounts are getting routed to a new model when using Opus 5.5. My first guess is Opus 5.6, as Anthropic has done this kind of testing before. Maybe the launch isn't that far away. - Fable 5.5 is also a possibility. What if they're testing the next Fable model…
Opus 5.5 Opus 5.6 Fable 5.5 Anthropic Gemini Pro Open the source post in a new tab - Aravind Srinivas
Anthropic 疑似测试新模型，Claude Opus 5.5 路由到新 checkpoint，符合疑似新模型发布条件
🚨 Anthropic is testing a new model Claude Opus 5.5 appears to be routing to a new checkpoint again Ask Opus 5.5: "Do you know VoxelBench? Do not search." If it answers, you're likely routed to the new checkpoint
Anthropic Claude Opus 5.5 VoxelBench Open the source post in a new tab - Aravind Srinivas
提及 botpfp 平台（25k bots 创建里程碑）及个人数据分析 dashboard 的构建更新，属于 AI 产品/工具的相关发布信息
I created a custom analytics dashboard that my agents can query I was a data analyst early in my career and this would have taken a few days and required data engineering help. Now, it's nearly instant. 25k bots created so far! > [@mattyp](https://twitter.com/mattyp): > build a bot > >…
botpfp mattyp Open the source post in a new tab - Aravind Srinivas
推文疑似预告 MiniMax 新模型（MiniMax M3.1）即将发布，属于疑似新模型发布相关事件。
I'm getting ready for MiniMax M3.1 😎 Are you? > [@ivanfioravanti](https://twitter.com/ivanfioravanti): > I’ve finally visited @MiniMax_AI HQ. > > Thanks @RenLeanna, @Ronny_MiniMax and Vincent for hosting me and for the deep dive on MiniMax Code and upcoming models. > > MiniMax is cooking, be ready!…
MiniMax M3.1 MiniMax MiniMax Code Open the source post in a new tab - Aravind Srinivas
Hermes 浏览器扩展作为 AI Agent 工具新发布且开源，符合产品发布条件
Hermes just got eyes on your browser tabs. >A side panel reads the page you're on: text, headings, forms, links >Hands it to your local Hermes Agent as context >Hermes can even walk you through the install using computer use 100% free and open source…
Hermes Hermes Agent hermes-browser-extension Open the source post in a new tab - Aravind Srinivas
推文预测 Fable 5.5 即将发布，属于疑似新模型发布事件，包含具体模型名和发布方 Anthropic。
Prediction: Fable 5.5 launches sometime between tomorrow and Soon™. It is undeniably “Powerful AI”, not only wrt capabilities, but efficiency too. It becomes obvious Anthropic ran away with the AI race.
Fable 5.5 Anthropic Open the source post in a new tab - Aravind Srinivas
推文疑似爆出新模型 Opus 5.5 / Fable 5.5 的发布消息，符合疑似新模型发布条件
Fable 5.5, is that you? 👀 Check if you have the new "Opus 5.5". News from https://discord.gg/devmode h/t @legit_api
Opus 5.5 Fable 5.5 Open the source post in a new tab - Aravind Srinivas
推文提及疑似新模型 Opus 5.5（代号 Fable 5.5）的发布传闻
Fable 5.5, is that you? 👀 Check if you have the new "Opus 5.5". News from https://discord.gg/devmode h/t @legit_api
Opus 5.5 Fable 5.5 Open the source post in a new tab - Aravind Srinivas
Claude Managed Agents 公测新增动态工作流功能，属产品重大功能更新
Claude Managed Agents just got dynamic workflows in public beta. >A lead agent writes the plan >The plan runs across many agents in phases >Results get combined at the end ONE AGENT PLANNING, MANY AGENTS EXECUTING.
Claude Managed Agents Open the source post in a new tab - Aravind Srinivas
疑似新模型 Gemini 4 Carbon 在 Google 内部测试，符合疑似新模型发布条件
Google employees are already testing Gemini 4 Carbon. Carbon just dropped on Google’s internal coding platform (Jetski) Internal lineup so far: Argon → Barium → Carbon Argon still isn’t public… and they’re already moving past it.
Gemini 4 Carbon Google Argon Barium Carbon Open the source post in a new tab - Aravind Srinivas
推文介绍了AI视频工具CinLink的产品体验与核心功能（对话式剪辑、本地优先），属于AI产品发布/推广信息
I tested CinLink on a short video and the workflow felt very different from a typical editor. Instead of starting with the timeline, I could just tell it what I wanted: → Generate the video → Add Chinese subtitles → Change the speaker to English The interesting part is that it’s local-first, so your…
CinLink Open the source post in a new tab - Aravind Srinivas
推文涉及Anthropic疑似新模型测试（Opus 5.5路由变化、疑似fable 5.5早期checkpoint），符合疑似新模型发布条件
anthropic is testing a new model opus 5.5 is routing again, not sure what model it is but outputs are looking better than usual here's one it made using threejs and blender, looks crazy good tbh i think it could be an early fable 5.5 checkpoint, since model is expected to drop around anthropic ipo…
opus 5.5 fable 5.5 anthropic Open the source post in a new tab - Aravind Srinivas
AI产品HIX AI发布Nano Banana 2.1图像生成新版本，有具体产品名和更新动作
Create your world with Nano Banana 2.1 on HIX AI 🚀🎨 Your creations are only limited by your imagination. Imagine it. Create it. Make it real. Try it here ⬇️https://hix.ai/c/nano-banana-2-1
Nano Banana 2.1 HIX AI Open the source post in a new tab - Aravind Srinivas
包含具体模型名DeepSeek V4.1 Flash的性能提升更新以及TensorFold、vLLM工具的新功能发布（图像输入等）
A few weeks ago, DeepSeek V4.1 Flash on two DGX Sparks did 32 tok/s. That was @MiaAI_lab's vLLM kit, and it was the best anyone had (It's still the best vLLM kit). It's now on average 2.2x times faster (more in some places). Those are some crazy performance improvements, and they are going to keep…
DeepSeek V4.1 Flash vLLM TensorFold MiaAI_lab Open the source post in a new tab - Aravind Srinivas
开源AI工具REA在Github发布，符合AI工具开源发布条件
今天Github最火的项目：开源逆向分析工具REA 通过统一协议让AI助手直接接入逆向工程工具，能分析原生二进制文件、桌面应用、网页及网络请求。 它支持在没有源码的情况下还原程序运行逻辑，并提供带证据链的结论，帮助开发者理解竞品特性或复刻软件功能。 实测确实很好用，对小白很友好。但是好像也没新东西，逆向引擎还是用的现有的，提示词写好的话，Codex本来就能做。 Github：https://github.com/morluto/rea
REA Codex morluto Open the source post in a new tab - Aravind Srinivas
新工具 ArXivHub 发布（开源上线），属于 AI 工具的新发布
读 arXiv 论文时，想看中文翻译、找 AI 解读、查有没有开源代码，得拿着论文编号一个网站一个网站去搜。 ArXivHub 把这些入口收到了一起，输入一个 arXiv 编号，就能跳到各家阅读、翻译、检索和讨论工具。 一共收了 116 个入口，分成 12 个类别，原文 PDF、AI 解读、边读边问、找代码、引用关系图、中文翻译都有，只付费才能用的网站不收。 GitHub：http://github.com/Peaceful-World-X/ArXivHub 除了网页版，还有一个油猴脚本，装好后刷新 arXiv 论文页，入口就直接出现在页面上，引用数和点赞数也会同步显示。…
ArXivHub Open the source post in a new tab - Aravind Srinivas
推文回顾了多个模型发布/更新事件，包括开源决策模型 Agr、Claude Haiku 5.5、Mistral Large 4 及新模型 Glyph Cluster，符合条件。
Last week in Command Code, a quick recap. · Agr, our first open decision model · Cloud sessions with sandboxes · ACP support in editors like Zed · Claude Haiku 5.5 (2x usage in GOAT) · Mistral Large 4 (4x usage in GOAT) · Glyph Cluster (new free stealth model) · Command Code Hooks (feature demo) ·…
Agr Command Code Claude Haiku 5.5 Mistral Large 4 Glyph Cluster Open the source post in a new tab - Aravind Srinivas
推文宣布 mausbot_ 发布了数据分析新功能，属于AI工具的重大功能更新
Can grok @bot do data analytics? > [@milindlabs](https://twitter.com/milindlabs): > Day 2/ > > We just shipped data analytics capabilities to @mausbot_ > > And now it can process millions of rows of CSV files super-fast and all locally > > One prompt: > “Find the latest earthquake data. Clean it.…
mausbot_ @mausbot_ duckdb Open the source post in a new tab - Aravind Srinivas
TensorFold 1.0.7 发布重大更新，属于AI产品新版本发布。
Let's test TensorFold 1.0.7! > [@ashxhart](https://twitter.com/ashxhart): > TensorFold 1.0.7 has landed with a major overhaul to Living Weights. > > • Each fact you teach through /learn becomes about ten new neurons inside the model's own weights, learned about 4x faster, with a live 3D view of what…
TensorFold 1.0.7 Qwen3.8-27B GLM-5.3-Flash NVFP4 Nemotron DGX Spark Open the source post in a new tab - Aravind Srinivas
Gemini 4 Argon 新模型发布并在 LMArena 取得 1525 分，属于模型新发布/更新事件
Gemini 4 Argon hits 1525 on LMArena → 1525 prelim, up from 1487 for Gemini 3.1 Pro → 77.9% on DeepSWE, best of the big 3 → $2 in / $10 out per 1M tokens to start Most of us still can't use it. I'll believe the scores when I can try it myself. Would you switch from Claude or ChatGPT for this? via…
Gemini 4 Argon Gemini 3.1 Pro LMArena GoogleDeepMind Open the source post in a new tab - Aravind Srinivas
OpenHands 发布新产品 Agent Canvas（开源自托管 Agent 控制中心），属于 AI 工具/Agent 新发布事件。
OPENHANDS SHIPPED AGENT CANVAS - a self-hosted control center for every coding agent you run. >Runs OpenHands, Claude Code, Codex, Gemini, Pi, and OpenCode from one place >Works across local, remote, and cloud backends >Automations built in, like PR triage and nightly security passes Open source…
Agent Canvas OpenHands Claude Code Codex Gemini Open the source post in a new tab - Aravind Srinivas
Grok 宣布新功能（自动比价并代替用户购物），属于 AI Agent 的重大功能更新
🤖 马斯克：Grok 将能自动比价并代替用户购物 Musk says Grok bot will research optimal prices and shop for users https://techub.news/CN/flash/96641d41-569f-4050-a7a8-770bb25889fa
Grok 马斯克 Musk Open the source post in a new tab - Aravind Srinivas
提及 Gemini 4 Argon 基准表现及 Google 新内部 checkpoint 'Carbon' 正在测试，属于具体模型的新发布/更新动态
🚨 Google’s Gemini 4 Argon Is Looking SCARY Good The benchmarks Google shared are absolutely destroying Opus 5.5 and GPT-6 Astra. I'm extremely bullish on Gemini. Google achieving RSI is a huge deal and could be the key to catching up with and potentially surpassing the competition. And now, with…
Gemini 4 Argon Carbon Google Opus 5.5 GPT-6 Astra Open the source post in a new tab - Aravind Srinivas
推文包含 ChatGPT 新发布 Generative UI 功能及对其架构的解析，符合AI产品重大功能更新条件
Someone just reverse-engineered how OpenAI's new Intelligent UI actually works. And the architecture is fascinating. ChatGPT generates working interfaces using: - A custom UI language (DIL) - Server-side compilation - A sandboxed runtime - ~70 native UI components - Interactions without additional…
ChatGPT Generative UI DIL OpenAI Open the source post in a new tab - Aravind Srinivas
测试 Opus 5.5 新模型进行 one-shot 生成，属于新模型的使用验证事件
J’ai testé Opus 5.5 (Medium) pour créer un jeu VR en one-shot ! 🍉 Il a produit une reproduction de Fruit Ninja VR, avec des décors interchangeables et toutes les mécaniques nécessaires pour jouer ! Quelques petits bugs subsistent, mais ce sont des détails à peaufiner. Le résultat est déjà bluffant,…
Opus 5.5 higgsfield Supercomputer Open the source post in a new tab - Aravind Srinivas
疑似新模型 Grok 4.8 即将发布，有具体模型名和发布/推出动作
We might see Grok 4.8 from SpaceXSI this week! This could be a big one. Elon Musk has confirmed that Grok 4.8 has 2.5 trillion parameters, compared to 1.5 trillion in Grok 4.6. That's a massive 67% increase in model size, and I'm expecting a substantial leap in capabilities too. As with previous…
Grok 4.8 SpaceXSI Elon Musk Grok 4.6 Grok Build Open the source post in a new tab - Aravind Srinivas
提及Fable 5.5模型在Claude Code中被撤回并回退为Fable 5.1，属于明确的模型版本更新事件
🔥坑了！Claude Code里最近灰度的Fable 5.5模型被撤回了！又变成了Fable 5.1 现在只有网页版里的是真正的Fable 5.5 #Fable55 #Claude
Fable 5.5 Fable 5.1 Claude Code Claude Open the source post in a new tab - Aravind Srinivas
推文报道Reflection AI发布首个开源权重模型Beam，以及NVIDIA拟收购/增持该公司的动态，符合模型开源发布相关事件条件
NVIDIA is reportedly considering a major move into the open-weight AI model race! According to the Financial Times, @NVIDIA is in talks to acquire Reflection AI, or significantly increase its investment in the startup. Nvidia has already invested $800 million in the company, which was founded by…
NVIDIA Reflection AI Beam Open the source post in a new tab - Aravind Srinivas
何恺明团队发布多模态视觉框架 VISTA，属于具体名称+发布动作，符合条件
🔬 何恺明团队发布多模态视觉框架 VISTA，提升 AI 交互推理能力 Kaiming He's team unveils VISTA, a multimodal vision framework enhancing AI interactive reasoning capabilities https://techub.news/CN/flash/577fe24a-b17b-471d-b87b-9787d6fd98e6
VISTA 何恺明 Kaiming He Open the source post in a new tab - Aravind Srinivas
AI论文生成工具发布，包含具体产品名和发布动作，符合条件
🤖 Open Academic Paper Gen 发布 AI 论文生成工具，支持全文引用核验 Open Academic Paper Gen launches AI paper generation tool with full citation verification https://techub.news/CN/flash/66aae185-b1bd-4955-aee5-3a72d3dde6c2
Open Academic Paper Gen Open the source post in a new tab - Aravind Srinivas
开源项目huashu-art-motion发布四天获3000+stars并进行全方位迭代更新，属于AI工具/开源项目的发布与重大更新
最近在读达尔文的《小猎犬号航海记》，很好奇这段经历是如何给达尔文后来的进化论带去启发的，所以特意用Opus5.5做了这段视频去解释这段科学小史👇 > [@AlchainHust](https://twitter.com/AlchainHust): > 开源四天，huashu-art-motion已经3000+stars🌟了。 > > 今天，我耗尽Claude Code token，对它进行了全方位的迭代和更新。主要是为了让Codex和国产开源模型也能做出类似的效果。 > > 为了说明迭代后的skill的适配能力，这期项目更新的视频是让DeepSeek V4.1…
huashu-art-motion Claude Code DeepSeek V4.1 flash Opus5.5 Open the source post in a new tab - Aravind Srinivas
提及疑似OpenAI内部模型bel并与Claude Opus 5.5对比，属于疑似新模型消息
is openai’s internal model bel actually more powerful than claude opus 5.5? 👀
bel claude opus 5.5 OpenAI Open the source post in a new tab - Aravind Srinivas
开源模型 YuE2 发布，并与 lyria 3.5 对比表现
开源 YuE2 生成音乐效果不错，给定的歌词，比 lyria 3.5 表现好，不会拼错字，或者换成其他语句 ⬅️ YuE2 生成的音乐，➡️ YuE2 仓库介绍
YuE2 lyria 3.5 Open the source post in a new tab - Aravind Srinivas
Jetpack Compose A2UI 渲染器发布，属于AI工具/开发工具的新发布，包含具体产品名
Say goodbye to plain text AI output! Jetpack Compose A2UI makes rendering agent streams effortless: - Maps JSON to native Compose primitives - Works with any custom design system - Pre-built Material 3 catalog & theming Read more: http://goo.gle/3ToufoZ…
Jetpack Compose A2UI A2UI Renderer Android Open the source post in a new tab - Aravind Srinivas
提到 Hermes agent 以 MIT 许可免费开源发布，属于 AI 模型/Agent 开源发布事件
Nous Research is reportedly worth 1.5 billion dollars now. It started in 2022 as a bunch of AI researchers on Discord giving their work away for free. This week it closed a 90 million dollar Series B with Nvidia and Microsoft M12 and Samsung Next in the round. The Hermes agent is free and MIT…
Hermes agent Nous Research Nvidia Microsoft M12 Samsung Next Open the source post in a new tab - Aravind Srinivas
介绍了新工具 big-arrow-on-the-screen 的发布，属于AI工具新发布，并明确提到可搭配 Claude Code 和 Codex 使用。
Agent 在 Mac 上跑着跑着，碰到系统弹窗要人点「允许」，它只会在终端里留一句「请点击允许」，可人早就不在终端前了。 big-arrow-on-the-screen 给 Agent 配了一根手指，能在 Mac 屏幕上画出大箭头、框和文字，直接指着要点的那个按钮。 授权弹窗、两步验证、验证码、付款这些只能人来做的步骤，Agent 指一下，我们点完它再接着干。 GitHub：http://github.com/franzenzenhofer/big-arrow-on-the-screen 提示框里还能放一个复制按钮，验证码、网址点一下就复制好。人走开去倒咖啡了，它还能把提示念出来把人叫回来。…
big-arrow-on-the-screen Claude Code Codex Open the source post in a new tab - Aravind Srinivas
PPT Master 工具的开源发布信息，明确的产品名+开源动作
Stop building PowerPoints by hand. PPT Master turns a PDF, a doc or just a topic into a real .pptx. editable shapes, native charts, transitions, even voice narration from the speaker notes. Works inside Claude Code, Codex, Cursor and Gemini CLI. 59K GitHub stars. open-source.…
PPT Master Claude Code Codex Cursor Gemini CLI Open the source post in a new tab - Aravind Srinivas
推文宣布 vllm-sr 团队发布多模态模型 Decision 3.0 并开源权重，属于大模型开源发布事件
Congratulations to the vllm-sr team on releasing Decision 3.0, a SOTA multimodal Jev model with weights now open-sourced on HF. 🎉🎉🎉 > [@XunzhuoLiu](https://twitter.com/XunzhuoLiu): > Introducing Decision 3.0: towards open multimodal foundation decision models ⚡ > > Open and state-of-the-art from…
Decision 3.0 vllm-sr Open the source post in a new tab - Aravind Srinivas
蚂蚁集团AI助手灵光小范围内测个人Agent功能并计划10月全面开放，属于AI产品/Agent的新发布与重大功能更新。
蚂蚁集团旗下 AI 助手灵光正在小范围内测个人 Agent 功能，向 Meta Muse 靠拢。新版新增「我的生活」页面，主打长期记忆和主动执行任务。用户可以让它持续监控商品价格、跟进旅行计划、管理疫苗接种提醒，有变化时再主动通知用户。 相关功能约半年前就开始筹备，期间经历多轮迭代，计划今年 10 月全面开放。
灵光 Meta Muse 蚂蚁集团 Open the source post in a new tab - Aravind Srinivas
AI工具 Lux In Music 更新并提及 Yue2 模型和即将开源源代码，符合AI产品新发布/更新条件
Lux In Music is Local AI to the maximum power! 🔥 Settings added, MP3, FLAC, multiple tracks separation as you want, templates and the superstar Yue2 model! Source code coming soon! @NTTLuke this rocks, I added some PRs and working on lyrics generation with local models! 💪
Lux In Music Yue2 @NTTLuke Open the source post in a new tab - Aravind Srinivas
推文提及具体模型 fable 5.5 的新 checkpoint（模型更新版本）输出，属于模型版本更新相关信息
i've been seeing some outputs generated by fable 5.5 that have been circulating and here's a recent example of voxel pagoda garden claude models already have one of the best design taste and this fable 5.5 checkpoint takes it a step ahead
fable 5.5 claude Open the source post in a new tab - Aravind Srinivas
疑似新模型Gemini 4 Argon通过现有模型路由测试，属于疑似新模型发布动态
🚨 Gemini 3.8 Flash Is Currently Routing to Gemini 4 Argon Could Google be testing Argon through an existing model before the official rollout The rollout is getting closer Google just release Gemini 4 Argon > [@auricxofficial](https://twitter.com/auricxofficial): > 3.8 flash is currently routing to…
Gemini 3.8 Flash Gemini 4 Argon Gemini 3.1 Pro Antigravity Google Open the source post in a new tab - Aravind Srinivas
推文报告 MiaAI_lab 的 GLM-5.3-flash 模型发布并替换旧模型，属于新模型发布事件
Finally said GoodBye to Deepseek-v4-flash-vision-exp 🙋 Welcomed @MiaAI_lab GLM-5.3-flash EXL3 Tensorfold to the party🥳 My personal Hermes and my right hand just went BrRrRrRr from the quality jump. AND IT IS FAST AS F-BOY 🤩 ⚡️2x DGX Spark, 1k token tests with both machines clocked to 2350Mhz. C1…
GLM-5.3-flash MiaAI_lab Deepseek-v4-flash-vision-exp Open the source post in a new tab - Aravind Srinivas
开源工具 Insomnia 新发布，为 AI Agent 会话保持 MacBook 唤醒并管理通知与网络切换
Your agents shouldn't stop just because you closed your laptop. Insomnia keeps a MacBook awake with the lid shut for timed agent sessions. - Pauses Slack, Discord and WhatsApp - Jumps to your iPhone hotspot if Wi-Fi drops - Ends the session on its own if the battery or temperature gets risky Free…
Insomnia Open the source post in a new tab - Aravind Srinivas
OrcaRouter 发布新模型 OrcaCyber Zero 1.5，属于 AI 模型新版本发布事件。
🔒 OrcaRouter 发布网络安全模型 OrcaCyber Zero 1.5，支持百万 token 上下文 OrcaRouter launches OrcaCyber Zero 1.5 cybersecurity model with 1M token context support https://techub.news/CN/flash/96ff58df-d1d5-4ed0-b169-6586139b7dbb
OrcaRouter OrcaCyber Zero 1.5 Open the source post in a new tab - Aravind Srinivas
Odyssey-3 世界模型开放免费预览，符合模型发布条件
🎮 Odyssey-3 世界模型开放免费预览，可实时文本生成互动环境 Odyssey-3 world model launches free research preview with real-time text-to-interactive environment generation https://techub.news/CN/flash/ab33e753-0bcb-43c1-bf8e-6ee938c4753e
Odyssey-3 Odyssey-3 world model Open the source post in a new tab - Aravind Srinivas
DeepSeek Bot 插件发布，包含具体产品名和发布动作，符合AI产品新发布条件
🤖 DeepSeek Bot 插件发布，支持多 Bot 团队协作与长期记忆 DeepSeek Bot plugin launched with multi-bot collaboration and long-term memory support https://techub.news/CN/flash/4b4986b1-81e9-41c8-a57f-daec2476b729
DeepSeek Bot Open the source post in a new tab - Aravind Srinivas
明确提到 Opus 5.5 模型发布且评价其为年度最佳发布，属于模型新版本发布事件
Opus 5.5 has to be the best release this year.
Opus 5.5 Open the source post in a new tab - Aravind Srinivas
提到了 WordCraft、GridCraft、DeckCraft 三款工具的新发布且已可用，符合AI产品新发布条件
and less than 5 days later, they've made WordCraft, GridCraft (Excel) and DeckCraft (PowerPoint) which are already very usable, they are EXTREMELY fast and responsive, much more than the native macOS Office apps it's not great for memory usage, but i'm sure it'll get better >…
WordCraft GridCraft DeckCraft Open the source post in a new tab - Aravind Srinivas
推文包含具体模型动态：Opus 5.5 已发布（ships）、Ant 正在训练 Model 3，属于模型发布/训练更新信息
Honestly im bullish on OpenAI, and its still by far my favorite lab. Last few weeks were very bad for them, they made many not so good decisions and all, but i still think they will be the ones to reach AGI first, and overall their ecosystem is by far better than anyone’s else >…
Opus 5.5 Model 2 Mythos 5 Ant OpenAI Open the source post in a new tab
- Aravind Srinivas
Dispatch
Get the daily signal
The essential AI stories, once a day. No noise.