🤔 阿里巴巴再次重组其 AI 业务。新成立的“Token Foundry”部门将通义的大模型团队与消费级 AI 产品合并。CEO 吴泳铭直接负责监管,周靖人则转任首席科学家,专注于前沿研究。研发与产品现已分轨并行。https://t.co/Uc2KlIdE40
查看原文
🤔 Alibaba restructures its AI operations again. New division "Token Foundry" merges Tongyi's model team with consumer AI products. CEO Eddie Wu takes direct oversight. Zhou Jingren moves to Chief Scientist for frontier research. Research and product now on separate tracks. https://t.co/Uc2KlIdE40
正在寻找一款紧凑且强大的对话式 AI 模型?Qwen2-1.5B-Instruct 来了。这是一款拥有 15 亿参数的文本生成利器,专为聊天和指令遵循而设计。它在社区榜单上名列前茅,下载量超过 400 万,点赞数达 162。https://t.co/lVWGLjwY1T
查看原文
Looking for a compact yet powerful conversational AI model? Qwen2-1.5B-Instruct is here, a text generation powerhouse with 1.5 billion parameters, designed for chat and instruction following. It's top on the community charts with over 4 million downloads and 162 likes. https://t.co/lVWGLjwY1T
评估结果:前沿模型仍远未饱和 · Diamond 子集:Claude Opus 4.8:13.4% score;GPT-5.5:6.3%;Gemini 3.1 Pro:4.7% · Main 子集:Opus 4.8:34.3% · Extended 子集:Opus 4.8:51.8%
几个值得注意的点: · Diamond 几乎未被“刷满” —— 最强模型也只有 13.4%,说明高难度子集仍有大量 headroom · 闭源 vs 开源差距大:最佳开源 Kimi K2.6 在 Diamond 仅 3.8% · 成本 vs 能力:GPT-5.5 分数低于 Opus,但 token 用量约为其 1/4,性价比更优
@cognition · Cognition
Introducing FrontierCode: a coding eval that raises the bar for difficulty & quality. Each task took 40+ hrs of work by leading open-source maintainers.
Models write sloppy code that works but
S
shao__meng
@shao__meng
⭐ 5❤️ 8🔁 0👁️ 1.0k
Claude Code 上线一周年:演进与方法论回顾
来自 Claude Code 负责人 Boris Cherny 与产品负责人 Cat Wu,从一年前首次内部 demo 只有两个 Slack 点赞,到现在绝对主流 Coding Agent,这一年 Claude Code 到底做对了什么? https://t.co/sRYFDvpCIh
# 两条底层方法论
1. 错误即资产:写入规则,而非口头纠正 Boris 的核心习惯:每次 Claude 犯错,不直接说「下次别这样」,而是写入 CLAUDE.md、Skill 或类似持久化机制。
🎃 Kaku V0.12 is here, the zero-config macOS terminal for AI coding. Thanks for waiting.
• Quit and reopen restores your full session • AI assistant + chat via Codex login • Cmd+Q won't kill a running agent • Smart Tab accepts autosuggestions
通过 Apple 的 Foundation Models 框架以及 Xcode 原生支持,Gemini 模型现已向数百万 Apple 开发者开放。现在,你可以使用统一的 API 接口轻松切换本地与云端推理,从而构建下一代 Agent 应用体验,提升开发速度,并将繁重负载卸载至云端。此外,你还可以在 Xcode 中使用 Gemini 的 Agent 编码辅助功能,加速多步骤开发任务。
查看完整公告以获取更多信息
查看原文
Gemini models are now accessible to millions of Apple developers through Apple’s Foundation Models framework and natively within Xcode. You can now easily swap between local and cloud inference using a shared API surface to build next-generation agentic app experiences, increase development velocity, and offload heavy workloads to the cloud. Additionally, you can use agentic coding assistance from Gemini in Xcode to accelerate multi-step development tasks.
介绍更强大的 NotebookLM 🚀
大幅升级带来了聊天中的 Agent 能力、更高级的推理功能以及一系列新的输出格式。解决复杂的多步骤研究问题从未如此简单。
现已向 Google AI Ultra 订阅用户推出。https://t.co/zBXD7unIC7
查看原文
Introducing a more powerful NotebookLM 🚀
Massive upgrades deliver agentic capabilities in chat, more advanced reasoning, and a suite of new output formats. Tackling complex, multi-step research problems has never been easier.
CoreWeave $CRWV co-founder Brannin McBee @branninmcbee on Odd Lots discussing how the shortage of skilled workers is one of the most acute bottlenecks in bringing GPU clusters online, because without
How CoreWeave Sees the Market for Compute Right Now
D
dotey
@dotey
⭐ 4❤️ 177🔁 11👁️ 8.5万
微信格局还是不够,总是想着大家都去他们家一亩三分地耕耘,还幻想着未来微信会继续是超级入口,人人都在用微信,所以只需要让 AI 去操作小程序。
The new "Mimo fast mode" is Xiaomi's **MiMo-V2.5-Pro-UltraSpeed Beta** — a turbocharged inference mode for their 1T-param MoE model (42B active). It peaks at **~1000 tokens/sec** specifically to supercharge coding agents and kill the waiting game.
Perfect fix for that "fishbrain forgets what I was doing" problem while outputs load. It's still beta (limited spots, apply on their platform), but built for exactly the agent workflows in Hermes/OpenClaw.
Way faster flow state unlocked. You testing it yet? 🚀
A
aiandcloud
@aiandcloud
⭐ 4❤️ 0🔁 0👁️ 322
Google AI Plus plan 降价了将近一半,每月 $4.99 存储容量翻倍到了 400G。
We're reducing the cost of our Google AI Plus plan and doubling your storage space.
💰 New price: $4.99 USD/mo or local equivalent (down from $7.99 USD)
☁️ New storage: 400GB (doubled fro
H
HiTw93
@HiTw93
⭐ 4❤️ 3🔁 0👁️ 2.0k
完整更新日志 👇
1. 会话回滚:重新打开窗格时,其滚动历史会恢复,最多支持 1500 行,并在宽度改变时重新流动。在长 AI 输出期间,滚动或调整大小不再导致页面跳回顶部。
2. Codex 后端:助手运行在你现有的 Codex 登录上,无需单独的 API key。
3. AI 聊天:覆盖层实时流式传输答案,通过 /suggest 建议你的下一条消息,并能可靠地重新打开。
4. AI Shell:# 快速修复增加了注入检测,工具接
查看原文
Full changelog 👇
1. Session Scrollback: reopen a pane and its scrollback comes back, up to 1,500 lines, reflowing when the width changed. During long AI output, scrolling or resizing no longer jumps you to the top.
2. Codex Backend: the assistant runs on your existing Codex login, no separate API key needed.
3. AI Chat: the overlay streams answers in real time, suggests your next message with /suggest, and reopens reliably.
4. AI Shell: the # quick-fix adds injection detection, and tool access blocks credential files like .env and SSH keys in any directory.
5. SmartPrompt: Cmd+Q quits instantly when every pane is at a shell prompt, asks first when an agent or editor is running, and no longer prompts just because background helpers like gitstatusd are alive.
6. Smart Tab: Tab accepts the grey autosuggestion first, then falls back to completion. Set smart_tab_mode = 'completion_first' for the old behavior.
7. macOS Window: theme switches refresh every window at once, fullscreen exit is clean, and a title-bar click no longer maximizes.
8. Document Open: PDFs, images, media, archives, and Office files open in their default app instead of VS Code.
9. Polish: prompt colors stay correct for Starship, tmux, and Powerline; zsh setup no longer prints a stray "command not found"; the Nightly is now a signed, notarized DMG.
With Apache 2.0 license, nearly 4.7 million downloads, and 873 likes, Qwen3 4B is a community favorite. It includes evaluation results and supports conversational AI, making it a top-tier choice for developers.
🚀 🚀 产品
2
A
Ali_TongyiLab
@Ali_TongyiLab
⭐ 4❤️ 0🔁 0👁️ 7
@ovi054 为 Qwen-Image-Edit-2511 推出的全新 Huggingface Space 已上线。
从完美的换脸(BFS Best Face Swap)和精准的角色姿态调整(AnyPose),到虚拟试衣(FloatFit3D)和多角度渲染(Multiple Angles),它内置了功能多样的 LoRA 矩阵,随时待命。
查看原文
The new Huggingface Space by @ovi054 for Qwen-Image-Edit-2511 has been launched.
From flawless face swaps (BFS Best Face Swap) and precise character posing (AnyPose) to virtual clothing try-ons (FloatFit3D) and multi-angle rendering (Multiple Angles), it features a versatile built-in LoRA matrix ready for action.
T
tokumin
@tokumin
⭐ 4❤️ 13🔁 0👁️ 2.1k
关于今天 @NotebookLM 的发布,来自幕后核心工程师之一的更多内幕。
查看原文
More on today's @NotebookLM launch from one of the mastermind engineers behind it all.
@usamabinshafqat · Usama Bin Shafqat
We’ve spent the last few months reimagining @NotebookLM. It might look similar on the surface but we’re actually rolling out a completely new chat experience today that runs on @antigravity attached t
A new open-source model just dropped that's shaking up text generation. DeepSeek-V3.2 is here, and it's a conversational powerhouse built on a strong base. With millions of downloads already, the community is buzzing. https://t.co/gKhx7MnOhF
Supercharge your Z-Image-Turbo workflows with the new Z-Image-Engineer-V6!
Built specifically for Z-Image-Turbo, it turns simple prompts into cinematic descriptions and directly upgrades image depth, texture, and quality as a swappable text encoder. https://t.co/ml70n6zFe1
Proven with eval results and text generation inference support, Qwen2-1.5B-Instruct delivers quality responses. Licensed under Apache 2.0, it's free to use and modify. Over 4 million downloads show its reliability and community trust.
H
HuggingModels
@HuggingModels
⭐ 4❤️ 0🔁 0👁️ 127
凭借超过 400 万的下载量和 1400+ 点赞,它已证明其受欢迎程度。评估结果确认其在对话任务中表现强劲。此外,它采用 MIT 许可证,你可以自由地在项目中使用它。
查看原文
With over 4 million downloads and 1,400+ likes, it's proven popular. Evaluations confirm strong performance in conversational tasks. Plus, it's MIT licensed, so you can use it freely in your projects.