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 10
Saturday- Aravind Srinivas
发布了具体产品 ODS（Osmantic 的私有 AI 服务器工具），附 GitHub 链接，属于 AI 工具发布
Turns a PC, Mac, or Linux box into a private AI server. https://github.com/Osmantic/ODS
ODS Osmantic Open the source post in a new tab - Aravind Srinivas
推文宣布kindlingai引擎开启GLM 5.3全量构建的beta测试，包含具体模型名与测试开放事件
We're opening up to testing today! Hop on the kindlingai discord if you want to pick up a TP=3 or TP=4 GLM 5.3 full build in a few hours and try it out. Been a few days of shaking out setup bugs in the brand-new engine. https://discord.gg/M7XTrRJW3 > [@mmastrac](https://twitter.com/mmastrac): >…
GLM 5.3 kindlingai kindlingai engine Open the source post in a new tab - Aravind Srinivas
微软发布了新AI模型 Microsoft-Decision-1，符合新模型发布条件
Microsoft has introduced Microsoft-Decision-1, a new AI model designed for fast and accurate decision-making. The model reportedly outperforms traditional LLMs and specialized decision models in both speed and quality on structured decision tasks. Microsoft is already testing it internally for…
Microsoft-Decision-1 Microsoft Open the source post in a new tab - Aravind Srinivas
腾讯发布开源模型 Youtu-Parsing-Omni（5B），包含明确的模型名与开源发布事件
Youtu-Parsing-Omni is now available as a compact 5B model for unified structured parsing. 🤖 https://modelscope.ai/models/tencent-community/Youtu-Parsing-Omni 🏆 Reaches 96.96 on OmniDocBench v1.6 and 75.08 on OmniParsingBench, ranking first among open-weight models on the latter. 🧩 A shared omni…
Youtu-Parsing-Omni 腾讯 Tencent Open the source post in a new tab - Aravind Srinivas
oMLX 0.7.1.dev1 发布，包含新功能与多模型推理加速更新
oMLX 0.7.1 dev1... Look at this GLM-5.3-Flash decode on M3 Ultra: 31.3 -> 40.6 tok/s (+30%). 🔥 > [@jundotkim](https://twitter.com/jundotkim): > oMLX 0.7.1.dev1 is out! > This dev release adds decision models (Clef, OpenJev), batched prefill for concurrent requests, a refreshed web UI, and faster…
oMLX 0.7.1.dev1 oMLX GLM-5.3-Flash Qwen3.8 Qwen3.6 Open the source post in a new tab - Aravind Srinivas
Claude Code Projects 正式发布并向 Pro/Max 用户开放，属于 AI 产品新发布事件
Claude Code Project正式发布了，做了一个双语视频。Project的目标是替换架构师，你只管布置庞大的任务，Claude会自行规划如何完成。 > [@ClaudeDevs](https://twitter.com/ClaudeDevs): > We just let in every Pro and Max user from the Claude Code Projects waitlist! > > If you're new to Claude Code Projects, here's a 4 minute walkthrough to get you started:
Claude Code Project Claude Code Projects Claude Open the source post in a new tab - Aravind Srinivas
发布了免费的3D角色建模工具キット「きゃらくる/Charakuru」，有具体产品名和发布动作
二次元玩3D 人物比我们还是溜啊～ 可以做二次元人物了 好玩 > [@Dstudio_ai](https://twitter.com/Dstudio_ai): > Tripo P2.0 × Codexで1枚絵からキャラクターモデルを制作できる「きゃらくる/Charakuru」というキットを無料公開しました！ > なんと、 Tripo様より正式にスポンサードしていただくことになりました！！！ > > Tripoだけでキャラクターモデルを制作しようとすると、顔がガビガビだったり、動きがおかしかったり、テクスチャが汚かったりしてしまいます…！ > このキットはそれを解決します！ > >…
Tripo P2.0 Codex きゃらくる/Charakuru GPT-6.1 Sol GPT-6 Astra Open the source post in a new tab - Aravind Srinivas
Syren Video 正式上线，属于 AI 产品新发布
AI video creation is getting ridiculously easy. From a simple prompt to cinematic, agency-quality videos, with the ability to edit everything through chat. Syren Video looks seriously impressive. 🔥 > [@vriparbelli](https://twitter.com/vriparbelli): > Syren Video is now live and FREE to try! > >…
Syren Video Opus 5.5 Syren renderer Open the source post in a new tab - Aravind Srinivas
推文包含具体产品/模型名 Nano Banana 2.1，且在 HIX AI 平台上线推广，属于 AI 产品新发布
Banana? Nano Banana! Your imagination, your creation. Make it happen with Nano Banana 2.1 on HIX AI! 👉https://hix.ai/c/nano-banana-2-1
Nano Banana 2.1 HIX AI Open the source post in a new tab - Aravind Srinivas
Cloudflare开源发布Clef-Omni多模态模型，包含具体模型名与开源动作
Cloudflare 开源 Clef-Omni：看图、听录音，直接做决策。 做客服分流、录音质检或视频事件判断，可以先定义问题和选项，再把材料交给它。模型一次前向计算，就能返回每道题各选项的概率，省掉自由文本生成和输出解析。 比如处理一段客服录音，可以同时问： • 应该转给哪个部门？从预设部门中选择。 • 有多紧急？按你定义的等级评分。 • 是否涉及服务故障？返回“是”的概率。 它支持文本、JSON、图像、音频和视频输入，基于 Qwen3-Omni-30B-A3B-Instruct 后训练，API 兼容 Jev / SystemOne，采用 Apache-2.0 协议。…
Cloudflare Clef-Omni Qwen3-Omni-30B-A3B-Instruct Open the source post in a new tab - Aravind Srinivas
引用的推文宣布ChatGPT订阅新增Devin订阅，属于产品功能更新事件
You guys are going hard on Tibo. I have never seen a post get 2 community notes at the same time. > [@thsottiaux](https://twitter.com/thsottiaux): > Your ChatGPT subscription is now also a Devin subscription
ChatGPT Devin thsottiaux Open the source post in a new tab - Aravind Srinivas
Midjourney 宣布推出 MCP 并招募测试者，属于 AI 产品/工具的新发布消息
Midjourney is launching an MCP! Apply to be a tester in the thread below. You'll need to explain what you would use it for. > [@midjourney](https://twitter.com/midjourney): > we want to start testing a Midjourney MCP with a limited community. the intention is to find creative and technical people to…
Midjourney Midjourney MCP Open the source post in a new tab - Aravind Srinivas
新工具 Obsidian Mind 发布，将 Obsidian 知识库变成 AI 编程助手的跨会话记忆，属于 AI 工具新发布
每次开新会话，都要给 AI 重讲一遍项目背景？ Obsidian Mind 把 Obsidian 知识库变成 AI 编程助手的跨会话记忆。适合长期用 Agent 做项目的人：把目标、决策和踩坑记录留下来，下次开工继续用。 按项目文档，这套流程可以这样跑： • /om-standup：开工前加载目标、进行中的项目、待办和近期 Git 变更。 • /om-dump：把会议内容说给 Agent，整理成人物笔记、决策记录和项目待办。 • /om-wrap-up：收尾时检查笔记链接、更新索引，补上遗漏的工作成果。 长期记忆保存在知识库文件里，可以用 Obsidian 查看、编辑，用 Git…
Obsidian Mind Obsidian Claude Code Codex CLI Gemini CLI Open the source post in a new tab - Aravind Srinivas
sparkDash 2.30 版本发布，属于AI/开发工具的重大功能更新，包含具体产品名与版本号及大量新功能说明
The latest sparkDash is AMAZING... if you have Sparks, this realtime dashboard is a MUST install... > [@MiaAI_lab](https://twitter.com/MiaAI_lab): > sparkDash 2.30 is out ⚡️ > > Windows PCs as units > > Monitor any Windows PC with an Nvidia GPU over its built-in OpenSSH. That covers RTX Spark…
sparkDash 2.30 sparkDash MiaAI_lab Open the source post in a new tab - Aravind Srinivas
Reppo宣布新增agent任务独立评审评分功能并上线Virtuals Protocol，属于AI产品的新功能发布
Moody’s for AI inference. @reppo now lets developers submit an agent’s work to independent judges and get back scores and critiques, helping applications decide what to accept, reject or retry. It is built so model-based judges like Jev can plug in. Live on Virtuals Protocol. >…
reppo Jev Virtuals Protocol Open the source post in a new tab - Aravind Srinivas
DataFast 发布了新功能：多网站汇总的收入与流量卡片及 /overview 仪表盘，属于产品功能更新
Done! If you have 2+ websites, DataFast will now show a new card showing total revenue & traffic for your websites. When you click it, it opens a new /overview dashboard. Is it useful? Should it stay in DataFast? > [@chrissyinspace](https://twitter.com/chrissyinspace): > Idea for @marclou's…
DataFast Open the source post in a new tab - Aravind Srinivas
Cloudflare 开源了基于 Qwen3-Omni 微调的新模型 Clef-Omni，包含明确的模型名与开源发布事件
Clef-Omni：Cloudflare开源的多模态版Jev 基于 Qwen3-Omni 微调，把文字、表格、图片、音频甚至视频交给它，同时提出几个设定好选项的问题。Clef-Omni 会在单次运算中看完所有材料，直接给出每个选项对应的概率。在企业办公自动化、意图识别和分类等测试场景下表现出色。 模型：https://huggingface.co/Cloudflare/clef-omni
Clef-Omni Cloudflare Qwen3-Omni Open the source post in a new tab - Aravind Srinivas
马斯克披露 Tesla 的 Digital Optimus 游戏智能体在训练并具备进展，属于疑似新模型/系统进展消息
Join @TeslaSI if you’re interested in solving generalized real-world super intelligence! > [@elonmusk](https://twitter.com/elonmusk): > We’re making interesting progress here. > > Digital Optimus can play about halfway through the Diablo campaign so far just by looking at the screen like a human. >…
Digital Optimus Tesla Elon Musk Open the source post in a new tab - Aravind Srinivas
微软发布新模型 Microsoft-Decision-1，属于大模型新版本发布
the question all were asking: microsoft-decision-1 would've been 19th place in the Decision Index 0.3.1 (it would have been the best score for a 9B parameter model. sadly, it isn't open weights so that doesn't matter as much) > [@satyanadella](https://twitter.com/satyanadella): > Introducing…
Microsoft-Decision-1 Microsoft satyanadella Open the source post in a new tab - Aravind Srinivas
微软发布新模型 Microsoft-Decision-1，含具体模型名与发布动作，属于新模型发布
the question all were asking: microsoft-decision-1 would've been 19th place in the Decision Index 0.3.1 (it would have been the best score for a 9B parameter model. sadly, it isn't open weights so that doesn't matter as much) > [@satyanadella](https://twitter.com/satyanadella): > Introducing…
Microsoft-Decision-1 microsoft-decision-1 Microsoft satyanadella Open the source post in a new tab - Aravind Srinivas
提及具体模型 Seedance 2.5 的能力表现，属于大模型相关的版本动态
Seedance 2.5 makes this look way too real. 👀 The details, expressions and movement are on another level. AI video just keeps getting better.
Seedance 2.5 Open the source post in a new tab - Aravind Srinivas
Odyssey-3 世界模型正式发布，属于新模型发布事件
You can experience pretty much anything with Odyssey-3. > [@odysseyml](https://twitter.com/odysseyml): > Today we're launching Odyssey-3, the most powerful foundation world model yet. > > It sets a new state of the art on Physics-IQ, and powers robots, trains AIs, and generates interactive…
Odyssey-3 odysseyml Open the source post in a new tab - Aravind Srinivas
Rapid-MLX v0.16.0 发布并新增 TensorFold profiles 提速功能，同时提到多个模型名
TensorFold popping up everywhere 🤯 > [@rapidmlx](https://twitter.com/rapidmlx): > 🚀 Rapid-MLX v0.16.0 is here! Turn your Mac into an absolute local AI powerhouse. > > Introducing TensorFold profiles with massive speedups: 🔥 DeepSeek V4 Flash: up to 2.1× throughput > 🧠 Bonsai 2 27B: +42% > ✨…
TensorFold Rapid-MLX v0.16.0 Rapid-MLX DeepSeek V4 Flash Bonsai 2 27B Open the source post in a new tab - Aravind Srinivas
介绍Charakuru工具包（Tripo P2.0 × Codex）的免费公开发布，属于AI工具的新发布
Charakuru 这个工具包让我震惊！ 内置了Blender和Tripo3D的工作流，开箱即用 1，Codex 生图+调用 Tripo 3D 生成 3D模型 2，Codex 控制 Blender 处理骨骼、头发和衣服物理 3，最后做出 VRM 虚拟形象，导出模型 大大节省做3D游戏的成本 地址放评论区👇 > [@Dstudio_ai](https://twitter.com/Dstudio_ai): > Tripo P2.0 × Codexで1枚絵からキャラクターモデルを制作できる「きゃらくる/Charakuru」というキットを無料公開しました！ > なんと、…
Charakuru きゃらくる Tripo P2.0 Codex Tripo Open the source post in a new tab - Aravind Srinivas
疑似新模型 Kimi K3.1 即将发布的爆料信息，符合疑似新模型发布条件
🚨 Kimi K3.1 Leaks: Drops soon > Kimi K3.1 is coming mid-to-late October > K3.1 is expected to improve long-horizon agent performance > Better post-training and computer-use capabilities > More real-world task data to improve agent intelligence > K3 remains the flagship, while K2.8 Preview focuses…
Kimi K3.1 K3.1 K2.8 Preview K3 K2.8 Open the source post in a new tab - Aravind Srinivas
Vivix 发布实时交互视频模型 W1 并上线 API，含具体产品名与发布动作
Vivix just put real-time AI video on its API at $0.003 per second. Here's what you need to know. The model is called Vivix W1. It's a real-time, streaming interactive video model, so video keeps generating as an experience unfolds. A creator on X, Zoya Khan, flagged the price on Oct 10, and her post…
Vivix W1 Vivix Seedance 2.5 Open the source post in a new tab - Aravind Srinivas
提及 Qwen 即将发布新模型，属于疑似新模型发布事件
Incoming new Qwen models! 🔥 > [@QwenDevs](https://twitter.com/QwenDevs): > big or small?
Qwen Open the source post in a new tab - Aravind Srinivas
AI监控工具sparkDash发布2.30新版本，包含多项新功能和修复
sparkDash 2.30! It rocks! > [@MiaAI_lab](https://twitter.com/MiaAI_lab): > sparkDash 2.30 is out ⚡️ > > Windows PCs as units > > Monitor any Windows PC with an Nvidia GPU over its built-in OpenSSH. That covers RTX Spark machines, which run Windows, and any other Windows box with an Nvidia card. > >…
sparkDash 2.30 sparkDash Open the source post in a new tab - Aravind Srinivas
疑似OpenAI新模型 gpt-rosalind-discovery 被发现出现在API定价页面，符合疑似新模型发布条件
New OpenAI model, and nobody announced it: gpt-rosalind-discovery. Same $5 price as the research one. Rosalind is their life sciences line. Drug discovery would be huge. For now it's only a guess. > [@testingcatalog](https://twitter.com/testingcatalog): > OPENAI 🔥: A new gpt-rosalind-discovery…
gpt-rosalind-discovery gpt-rosalind-research OpenAI Open the source post in a new tab - Aravind Srinivas
独立开发者发布了 ImageFlow 的 relight 功能，属于 AI 产品/工具的新发布，虽然涉及对比 Adobe 但核心是产品发布
In the past decade, Adobe has been mostly focusing on awkward licensing models to increase their cash flow. It's good to finally see people beating Adobe. Even vibe coded image editors do some things better today and they load instantly. Adobe software is so heavy. >…
ImageFlow ImageFlow.dev Adobe LightRoom Open the source post in a new tab - Aravind Srinivas
发布了 MiniMax-H3-Turbo 模型，是音频视频生成模型的开源/新发布信息
Introducing MiniMax-H3-Turbo from Wuli-Art. MiniMax-H3 audio-video generation now runs in as few as four steps. 🤖 https://modelscope.ai/models/Wuli-Art/MiniMax-H3-Turbo ⚡ Choose from 4-step, 6-step, and 8-step variants for faster audio-video generation. 🎬 Supports text-to-video, first-frame and…
MiniMax-H3-Turbo MiniMax-H3 Wuli-Art Open the source post in a new tab - Aravind Srinivas
Qwen官方暗示将发布大小两个新模型，属于疑似新模型发布事件，包含具体模型名Qwen
Big and small, both coming. Qwen made it official. My bet: Qwen 4 27B and Qwen 4 Max. That's only my guess, no names have been confirmed. Can't wait to try the small one. > [@QwenDevs](https://twitter.com/QwenDevs): > alright guys, let's do both!
Qwen Qwen 4 27B Qwen 4 Max QwenDevs Open the source post in a new tab - Aravind Srinivas
Google 对 Gemini API 模型进行了更新/弃用及自动路由调整，属于模型更新事件。
Google just reshuffled the Gemini API. Deprecated: • Gemini 3.7 Flash • Gemini 3.5 Flash • Deep Research Pro → shuts down Oct 23 Auto-rerouting • 3.7 Flash → auto-routed to 3.8 Flash • 3.5 Flash → auto-routed to 3.6 Flash Update your model strings.
Google Gemini API Gemini 3.7 Flash Gemini 3.5 Flash Deep Research Pro Open the source post in a new tab - Aravind Srinivas
推文包含具体模型名 GPT-6.1 Astra 及疑似新模型相关事件（安全测试未通过、仍在训练其他模型），属于疑似新模型发布动态
Yesterday we focused on Anthropic rumors, but make no mistake, OpenAI isn't just sitting around doing nothing. The GPT-6.1 Astra variant failed its safety tests, but the team is responding incredibly quickly and professionally, and this hasn't affected the training of other models. Don't mistake…
OpenAI GPT-6.1 Astra Anthropic Open the source post in a new tab - Aravind Srinivas
推文提到疑似新模型 Gemini 4 argon 出现在 Antigravity 中，且 Google 将于 11 月 2 日退役旧模型，符合新模型疑似发布条件
gemini 4 argon is literally showing up in antigravity, context window options are appearing, and google is preparing to retire older models on november 2 yet most of us still can't use argon 😭 google, why are we getting the migration before the model?
gemini 4 argon antigravity google Open the source post in a new tab - Aravind Srinivas
Google 发布全新免费 AI 会议笔记应用 AI Edge Foresight（基于 EmbeddingGemma 2），属于 AI 产品新发布事件
100 percent free and fully local. That is the new Google AI notetaker for Mac called AI Edge Foresight. It runs on EmbeddingGemma 2 and the demo shows 1.73 million tokens saved in a single meeting. Every premium feature works with no subscription. You get live enhanced notes and an AI summary and a…
AI Edge Foresight EmbeddingGemma 2 Embedding Gemma 2 Google Open the source post in a new tab - Aravind Srinivas
预告 Qwen 4 新模型即将发布，属于疑似新模型发布事件
Let's get ready! Qwen 4 is coming! > [@QwenDevs](https://twitter.com/QwenDevs): > big or small?
Qwen 4 QwenDevs Open the source post in a new tab - Aravind Srinivas
OpenAI 在 Codex 中发布新功能 composer predictions，属于产品新发布/重大功能更新
OpenAIDevs 宣布 composer predictions，可依据对话和用户表达方式预测下一条消息。 Pro 套餐内使用不消耗 usage！ 不说了，赶紧去codex里玩儿一下 > [@OpenAIDevs](https://twitter.com/OpenAIDevs): > Now in beta: composer predictions in Codex for Pro users. > > Codex can now suggest your next message based on your conversation and how you talk to it. >…
composer predictions Codex OpenAI Open the source post in a new tab - Aravind Srinivas
Higgsfield Katana 新上线，作为 AI 工具通过 Higgsfield MCP 集成到 Claude，符合 AI 产品新发布条件
Higgsfield Katana is live inside Claude, running through the Higgsfield MCP. Upload a reference and it spits out editable motion graphics, product launch videos, aura-farming edits. Editable motion graphics is the part worth staring at.
Higgsfield Katana Claude Higgsfield MCP Higgsfield Open the source post in a new tab - Aravind Srinivas
Dreamina AI 宣布其首个专业影视内容品牌 Dreamina Originals 项目陆续上线，属于AI产品的新发布动态
Dreamina Originals Projects Are Rolling Out As Dreamina AI's first professional film and series content label, Dreamina Originals provides creators with computing power, technical support, AI creative tools, and premium distribution resources. Recently, projects under Dreamina Originals have been…
Dreamina Originals Dreamina AI Dreamina Open the source post in a new tab - Aravind Srinivas
微软发布新模型 Microsoft-Decision-1，属于AI模型新发布
Microsoft just introduced an entirely new class of AI model, and the performance numbers are pretty wild! Microsoft-Decision-1 is a specialized model designed to make decisions instead of generating text. Think AI agent routing, classification, evaluating responses, choosing the next action, and…
Microsoft-Decision-1 Microsoft Qwen3.5-9B GPT-6 Sol Open the source post in a new tab - Aravind Srinivas
Vivix 发布实时交互视频模型 W1 并上线 API，属于新模型发布
An hour of live AI video generation now costs $10.80 At $0.003 a second, Vivix W1 generates a full minute for just 18 cents. That price matters because of what W1 is. It's a real-time interactive video model. Through the API, it keeps generating video while the experience is happening, and you can…
Vivix W1 W1 Vivix Seedance 2.5 Kling 3.0 Open the source post in a new tab - Aravind Srinivas
推文包含具体模型名及疑似新模型/训练发布相关信息（Claude 6、Model 2、Fable 5.5 疑似推出与训练计划）
🚨 I really think Anthropic might be building something far more dangerous to OpenAI than another Claude model. I've been hearing some wild rumours about Anthropic's next training effort. → A massive new training run could begin in the coming weeks, potentially laying the groundwork for Claude 6. →…
Anthropic OpenAI Claude Claude 6 Model 2 Open the source post in a new tab - Aravind Srinivas
Pine AI 发布面向 AI Agent 的云电脑产品 Pine Computer，属于 AI 产品/Agent 新发布
Pine AI 推出专为 AI Agent 设计的云电脑 Pine Computer。它整合了云端虚拟机、AI 模型和操作工具。开发者通过 API 下达任务，AI 就能在云端操作浏览器、软件和文件，不必使用用户自己的电脑。 Pine 的核心改进是让电脑主动告诉 AI 发生了什么变化。传统视觉式 Agent 通常需要反复截图，才能判断点击后的结果。Pine 则让浏览器和操作系统直接向执行任务的 AI 模型反馈变化，减少反复截图和检查页面的次数。 Pine 宣称，内部测试中，新方案执行任务的速度达到传统方案的 2 至 5 倍。在 SaaS-Bench v1.1 测试中，Pine 的检查点得分达到…
Pine AI Pine Computer Claude Code Codex Open the source post in a new tab - Aravind Srinivas
作者为 TensorFold 发布新量化框架 EAQ，并针对 GLM 5.3 Flash 模型，属于 AI 模型相关新发布
I've been building something new for TensorFold. It's not just another quant. It’s a quantization framework designed around how TensorFold actually runs models. I'm calling it Engine-Aware Quantization, or EAQ. First target: GLM 5.3 Flash. Early results vs my current public quant on the same…
TensorFold EAQ Engine-Aware Quantization GLM 5.3 Flash Open the source post in a new tab - Aravind Srinivas
蚂蚁集团旗下inclusionAI开源PolyOCR-VENUS OCR基础模型系列，属于大模型开源发布
Ant Group’s inclusionAI just open-sourced PolyOCR-Venus, a family of 2B and 9B OCR foundation models built on the Qwen3.5 vision-language backbone. One model family brings together text recognition, localization, document parsing, information extraction, visual text understanding, and OCR reasoning.…
inclusionAI PolyOCR-Venus PolyOCR-9B Qwen3.5 Ant Group Open the source post in a new tab - Aravind Srinivas
涉及 Qwen 4 疑似新模型发布及开源（open-weight）的泄露与官方回应，符合新模型发布/开源事件条件
Qwen Developers just teased Qwen 4 and said they'll release two open-weight models. Here's what you need to know. On Oct 10, 2026, the official @QwenDevs account asked its followers "big or small?" about Qwen 4. About 5 hours later, a post from @0x0SojalSec went viral, claiming Qwen 4 is coming and…
Qwen 4 Qwen4-27B Qwen4-Max Qwen Developers Alibaba Open the source post in a new tab - Aravind Srinivas
推文包含具体的未发布模型名（Gemini 4 Argon、Gemini 4 Carbon）及疑似推出/测试事件
Everything looks ready for Gemini 4 Argon now Argon is all setup in Antigravity and all old models are leaving soon Google please just release this thing already > [@LuminaBench](https://twitter.com/LuminaBench): > Google are already testing another Gemini 4 model called "Carbon" in Jetski > >…
Gemini 4 Argon Argon Gemini 4 Carbon Carbon Google Open the source post in a new tab - Aravind Srinivas
涉及 Gemini 4 内部测试版本及后续模型更新的具体报道
Google hasn't even rolled out Gemini 4 Argon to the public yet, and its engineers are already testing a MUCH more powerful successor! According to Business Insider, @GoogleDeepMind has been testing several Gemini 4 variants internally, codenamed Argon, Barium, and Carbon. The latest, Carbon, is…
Gemini 4 Argon Google GoogleDeepMind Barium Carbon Open the source post in a new tab - Aravind Srinivas
YouWare 发布了 Website 模板新功能，属于 AI 产品重大功能更新。
New in YouWare: Website templates. Right next to Presentation and Report on the home screen, there’s now a Website tab: 15 web page templates. Pick a look → pick any model → get a real HTML page. Then share it and work on it with your team. Start fast: http://youware.com
YouWare Open the source post in a new tab - Aravind Srinivas
推文报道Google内部疑似测试新模型Gemini 4（argon/carbon/barium-b），属于疑似新模型发布事件
google got opus 5.5 level model reportedly employees are testing newer gemini 4 version called carbon and it's comparable to opus 5.5 earlier argon versions were compared to opus 5 but looks like team been cooking with newer checkpoint > argon is the gemini 4 pro model > barium-b is the internal…
Gemini 4 carbon argon barium-b Google Open the source post in a new tab - Aravind Srinivas
宣布 vLLM Semantic Router 新功能 System One Auto 发布，涉及 Kai 0.6B 与 Vega 27B 模型的路由调度更新
No single decision model wins. Introducing System One Auto in vLLM Semantic Router. Kai 0.6B answers first; Vega 27B only when needed. Public JevBench requests: -54% est. cost, -20% mean latency vs Vega-only +16.45 pts accuracy vs Kai alone #Jev #vLLM
System One Auto vLLM Semantic Router Kai 0.6B Vega 27B JevBench Open the source post in a new tab - Aravind Srinivas
疑似新模型发布：Memory4 Dream 模型出现，疑似基于 GPT-6 Luna
Better memory lets go! > [@Lentils80](https://twitter.com/Lentils80): > A new GPT memory consolidation model, called Memory4 Dream (ID: memory4-dream), has appeared a few hours ago and it seems to be based on GPT-6 Luna.
Memory4 Dream memory4-dream GPT-6 Luna Open the source post in a new tab - Aravind Srinivas
包含Jev模型发布事件，提到launch a model
Huge congrats to @CompleteSkeptic and the @typesafeai team! I was an angel investor in the seed round, and I remember when they were trying to raise their Series A--almost every investor passed because they didn't get it, "the team is too eccentric," and "non-LLM approaches are a dead end." After…
Jev Typesafe a16z Open the source post in a new tab - Aravind Srinivas
作者开源发布了新工具 work-humanizer，属于AI工具的开源发布
One more test of the work-humanizer skill. In a business context it's critical that AI doesn't: - Invent facts - Makes commitments - Changes the meaning Beats every other method. Free, open source: https://github.com/phuryn/work-humanizer > [@PawelHuryn](https://twitter.com/PawelHuryn): > I've been…
work-humanizer Emulate-1 Pangram PawelHuryn Open the source post in a new tab - Aravind Srinivas
开源修图软件 Compositor 的发布信息，属于工具/产品的开源发布，且提到为 AI Agent 留接口
Photoshop 订阅一年下来不便宜，换成 GIMP 又总觉得哪哪儿都不顺手，面板和快捷键都得重新记。 Compositor 是一款 Mac 上的开源修图软件，作者以前用 Photoshop 做合成和后期，干脆照那套习惯自己写了一个，已经斩获 1.4 万多 Star！ 图层、蒙版、剪贴蒙版、调整图层都有，混合模式按 Photoshop 的顺序排，快捷键也是同一套，不顺手的还能自己改。 GitHub：http://github.com/robbietilton/Compositor 能直接打开 PSD 和 PSB…
Compositor GIMP Photoshop robbietilton Open the source post in a new tab - Aravind Srinivas
AI编程工具Amp宣布重大功能更新：支持Claude Pro/Max订阅接入，包含具体产品名和明确更新动作
AI 编程工具 Amp 宣布，现已支持 Claude Pro/Max 订阅，向所有用户免费开放接入。新建任务时选择「Claude Code」模式，绑定 Claude 账号即可。模型使用量直接计入原有订阅限额，无需再向 Amp 单独购买额度。 新模式通过 Claude Agent SDK 运行官方 Claude Code，同时保留 Amp 的云端电脑、任务共享、多人协作和 Agent 编排等功能。只有这个模式能使用 Claude 订阅，Amp 原有的 Medium、High、Ultra 模式不在其中。本地运行免费，云端电脑仍需付费。 Anthropic 此前曾限制第三方工具使用 Claude…
Amp Claude Pro/Max Claude Code Claude Agent SDK Anthropic Open the source post in a new tab - Aravind Srinivas
腾讯云开源了TeamAI工具，包含具体产品名和开源动作
🔥Tencent Cloud has open-sourced TeamAI, an internal tool that has made GitHub Trending. It’s a Git-based tool for sharing skills and team knowledge across AI agents. TeamAI puts team skills, rules, tool configs, and project knowledge in a shared Git repo, where they can be reviewed and versioned…
TeamAI Tencent Cloud WorkBuddy CodeBuddy Claude Code Open the source post in a new tab - Aravind Srinivas
宣布新模型JEV-27B-VL的发布与开源，属于模型开源发布事件
JEV-27B-VL brings accurate, calibrated visual decisions to long-context workflows. 🤖 https://modelscope.ai/models/AutoTrust-Labs/JEV-27B-VL 📄 https://modelscope.ai/papers/2512.16899 🏆 Ranks #1 among 20 models on the independent Decision Index Vision board, with 77.8% accuracy across 13,695…
JEV-27B-VL AutoTrust-Labs Qwen3.5 Open the source post in a new tab - Aravind Srinivas
推文涉及疑似新模型 Gemini 4 系列的发布消息，包含具体模型名和发布动作
Gemini 4 models might be coming next week gemini app just got upgrade with new thinking options for each model old models are also leaving antigravity in few days, so team might be preparing to add whole gemini 4 family as far as i know, argon is the pro model, so we could get: > gemini 4 > gemini 4…
Gemini 4 gemini 4 flash gemini 4 argon Gemini antigravity Open the source post in a new tab
- Aravind Srinivas
Dispatch
Get the daily signal
The essential AI stories, once a day. No noise.