Scroller Briefings - Xiaomi Officially Unveils the Flagship MiMo-V2 Series of Large Models and Claims “Hunter Alpha,” Promising Future Open-Source Release 🚀
编辑
目录
单词学习：![[2026031993303481.mp3]]
新闻播报：![[2026031999844225.mp3]]
小米正式发布 MiMo-V2 系列旗舰大模型并认领“Hunter Alpha”，承诺未来开源
简报：
- 今日凌晨，小米正式推出三款自研大模型——旗舰基座 MiMo-V2-Pro、全模态基座 Omni 以及语音合成 TTS，并认领了此前引发热议的神秘模型“Hunter Alpha”。
- 旗舰模型 MiMo-V2-Pro 拥有超 1T 总参数量，支持 1M 超长上下文，专为 Agent 工作场景打造，目前 API 定价仅为同级别模型的 1/5。
- MiMo-V2-Pro 已与金山办公生态实现框架级集成，原生支持多种主流文档格式，WPS 灵犀现已接入该模型。
- 同步上线的 MiMo-V2-Omni 支持音视频深度理解与复杂交互；MiMo-V2-TTS 支持多方言、多粒度情绪调节及高保真歌声合成。
- 此次发布的大模型已登录 Xiaomi miclaw 等多个平台，并联合五大 Agent 框架提供为期一周的限时免费体验。
- 小米 MiMo 大模型负责人罗福莉发文回顾研发历程，并承诺 MiMo-V2 系列模型在足够稳定后将会正式开源。
相关链接：
Xiaomi Officially Unveils the Flagship MiMo-V2 Series of Large Models and Claims “Hunter Alpha,” Promising Future Open-Source Release 🚀
Briefing:
- Early this morning, Xiaomi officially launched three in-house large models—the flagship foundation model MiMo-V2-Pro, the multimodal foundation model Omni, and the speech synthesis model TTS—and confirmed that it is behind the mysterious “Hunter Alpha” model that had previously sparked widespread discussion.
- The flagship model MiMo-V2-Pro features more than 1T total parameters, supports an ultra-long 1M context window, and is specifically designed for Agent-based workflows. Its current API pricing is only one-fifth that of comparable models.
- MiMo-V2-Pro has already achieved framework-level integration with the Kingsoft Office ecosystem, with native support for multiple mainstream document formats. WPS Lingxi has now integrated the model.
- Also launched alongside it, MiMo-V2-Omni supports deep audio and video understanding as well as complex interactions; MiMo-V2-TTS supports multiple dialects, fine-grained emotion control, and high-fidelity singing voice synthesis 🎤
- The newly released large models are now available on multiple platforms, including Xiaomi miclaw, and Xiaomi has partnered with five major Agent frameworks to offer a one-week limited-time free trial.
- Luo Fuli, head of Xiaomi’s MiMo large model team, published a post reviewing the R&D journey and promised that the MiMo-V2 series will be officially open-sourced once it is sufficiently stable 🔓
Related links:
- Xiaomi launches three in-house MiMo-V2 series models late at night: flagship Pro / multimodal foundation model Omni / speech synthesis TTS
- Xiaomi claims the mysterious top-tier model rumored online: Luo Fuli promises the MiMo-V2 series will be open-sourced
小米正式发布MiMo-V2系列大模型，负责人罗福莉承诺未来将开源
简报：
- 3月19日，小米正式发布三款专为Agent时代打造的自研大模型：旗舰基座MiMo-V2-Pro、全模态基座Omni以及语音合成模型TTS，并开放限时一周免费体验。
- 小米官方正式认领了此前在全球最大API聚合平台OpenRouter上引发热议并登顶调用榜的神秘模型“Hunter Alpha”，确认其为MiMo-V2系列。
- 小米MiMo大模型负责人罗福莉发布长文回顾了1T基础模型的研发与智能体框架的转型历程，并明确承诺当模型足够稳定时，新的MiMo-V2系列模型将会开源。
相关链接：
Xiaomi Officially Launches the MiMo-V2 Series of Large Models, with Lead Luo Fuli Promising Future Open-Source Release 🚀
Brief:
- On March 19, Xiaomi officially launched three self-developed large models built for the Agent era: the flagship foundation model MiMo-V2-Pro, the omni-modal foundation model Omni, and the speech synthesis model TTS, while also offering a one-week free trial for a limited time.
- Xiaomi officially confirmed that the mysterious model “Hunter Alpha,” which had previously sparked widespread discussion and topped the usage rankings on OpenRouter, the world’s largest API aggregation platform, is in fact part of the MiMo-V2 series.
- Luo Fuli, head of Xiaomi’s MiMo large model team, published a long post reviewing the development of the 1T base model and the transition toward an agent framework, and explicitly promised that the new MiMo-V2 series models will be open-sourced once they are sufficiently stable. 🤖✨
Related links:
- Xiaomi launches three self-developed MiMo-V2 series models late at night: flagship Pro / omni-modal foundation Omni / speech synthesis TTS
- Xiaomi claims the widely discussed mysterious top-tier model online: Luo Fuli promises the MiMo-V2 series will be open-sourced
小米发布三款新大模型并上线 Xiaomi MiMo Claw，开启限时免费体验
简报：
- 小米于3月19日推出 MiMo-V2-Pro、Omni 及 TTS 三款大模型，并宣布开启为期一周的限时免费体验。
- Xiaomi MiMo Studio 官网上线 Xiaomi MiMo Claw 私人助理，搭载最新旗舰及多模态理解模型，支持一键极速部署 OpenClaw。
- 该工具深度集成金山办公生态，内置 WebOffice 在线预览，支持 Word、Excel、PPT、PDF 等超 95% 的主流文档类型。
- 用户单次可免费体验 30 分钟，工具自带文件系统并支持网页图文抓取存储，退出体验后相关数据会自动销毁。
相关链接：
Xiaomi Releases Three New Large Models and Launches Xiaomi MiMo Claw, Opening a Limited-Time Free Trial 🚀
Brief:
- On March 19, Xiaomi launched three large models — MiMo-V2-Pro, Omni, and TTS — and announced a one-week limited-time free trial.
- The Xiaomi MiMo Studio official website has launched the Xiaomi MiMo Claw personal assistant, powered by the latest flagship and multimodal understanding models, with support for one-click rapid deployment of OpenClaw.
- The tool is deeply integrated with the Kingsoft Office ecosystem, featuring built-in WebOffice online preview and support for more than 95% of mainstream document formats, including Word, Excel, PPT, and PDF.
- Users can enjoy a free 30-minute trial per session. The tool comes with its own file system and supports web page text-and-image capture and storage. After exiting the trial, all related data will be automatically destroyed. 🤖📄
Related Links:
- Xiaomi MiMo Claw Opens Free Trial, Integrates Kingsoft Ecosystem and Supports One-Click OpenClaw Deployment
- Official Xiaomi MiMo Claw Trial Page
黄仁勋盛赞OpenClaw为下一个ChatGPT，英伟达GTC 2026发布企业级NemoClaw与新一代超算
简报：
- 北京时间2026年3月17日，英伟达CEO黄仁勋在GTC 2026大会上表示，AI智能体开源项目OpenClaw（龙虾）是“人类历史上规模最大、最受欢迎、最成功的开源项目”，并预言其绝对是“下一个ChatGPT”。
- 黄仁勋将OpenClaw定义为AI智能体计算机的操作系统，宣称其仅用几周时间就超越了Linux在过去30年取得的成就。
- 为解决OpenClaw直接操作电脑带来的潜在安全隐患，英伟达正式推出了基于该项目的企业级软件栈NemoClaw，通过内置的隔离沙箱和隐私路由器提供安全、可控且一键安装的AI智能体定制服务。
- 为支撑激增的智能体算力需求，英伟达发布了由7种芯片和5种机架系统组成的Vera Rubin超级计算机平台，该平台采用100%液冷设计，可将Token生成速率提升350倍。
- 黄仁勋在大会上提出了“Token工厂经济学”概念，指出数据中心将转变为Token生产工厂，并预言Token额度正成为硅谷新的招聘筹码，同时呼吁全球企业制定专属的OpenClaw战略。
相关链接：
Jensen Huang Praises OpenClaw as the Next ChatGPT; Nvidia Unveils Enterprise-Grade NemoClaw and a New Generation of Supercomputers at GTC 2026 🚀🦞
Briefing:
- On March 17, 2026, Beijing time, Nvidia CEO Jensen Huang said at the GTC 2026 conference that the open-source AI agent project OpenClaw (Lobster) is “the largest, most popular, and most successful open-source project in human history,” and predicted that it is definitely “the next ChatGPT.”
- Huang defined OpenClaw as the operating system for AI agent computers, claiming that it surpassed what Linux achieved over the past 30 years in just a matter of weeks.
- To address the potential security risks associated with OpenClaw directly operating computers, Nvidia officially launched NemoClaw, an enterprise-grade software stack built on the project. With built-in isolated sandboxes and privacy routers, it provides secure, controllable, and one-click AI agent customization services 🔒.
- To support the surging demand for agent computing power, Nvidia introduced the Vera Rubin supercomputing platform, consisting of seven types of chips and five rack-scale systems. The platform features a 100% liquid-cooled design and can increase token generation speed by 350 times.
- At the conference, Huang introduced the concept of “Token Factory Economics,” pointing out that data centers will transform into token production factories. He also predicted that token quotas are becoming Silicon Valley’s new hiring currency and called on companies worldwide to formulate their own OpenClaw strategies 💡.
Related link:
腾讯刘炽平回应微信Agent进展：潜力巨大面临挑战，上线暂无明确时间表
简报：
- 3月18日的腾讯控股2025年业绩沟通会上，腾讯总裁刘炽平透露，过去一段时间已在微信中尝试多项AI功能，为未来的微信Agent铺路。
- 刘炽平表示微信生态丰富且商业模式良好，微信Agent“可以做很多实际的事”并让合作方获益，但目前推出尚无特定时间表，正积极推进。
- 该项目推进过程中仍面临多重挑战，包括保护用户隐私安全、通用模型难以满足微信独有功能需求，以及庞大用户规模所需的极高推理能力。
- 据此前报道，微信正在研发一套独立的自有AI模型，已完成基础能力建设，预计将于2026年对外落地并接入小程序生态。
- 另有消息称，腾讯正在秘密开发一款连接数百万小程序的“绝密级”微信AI智能体，未来有望取代14亿月活用户手动完成外卖、网约车等任务的需求。
相关链接：
Tencent’s Martin Lau Responds to Progress on WeChat Agent: Huge Potential but Significant Challenges, No Clear Launch Timeline Yet 🚀🤖
Briefing:
- At Tencent Holdings’ 2025 earnings communication meeting on March 18, Tencent President Martin Lau revealed that the company has been testing a number of AI features within WeChat over the past period, laying the groundwork for a future WeChat Agent.
- Lau said that WeChat has a rich ecosystem and a strong business model. The WeChat Agent “can do many practical things” and benefit partners, but there is currently no specific timetable for its launch, though development is being actively advanced.
- The project still faces multiple challenges, including protecting user privacy and security, the difficulty of using general-purpose models to meet WeChat’s unique functional needs, and the extremely high inference capacity required to serve its massive user base.
- According to previous reports, WeChat is developing an independent in-house AI model. The foundational capabilities have already been completed, and the model is expected to be launched externally in 2026 and integrated into the Mini Program ecosystem.
- Other reports also claim that Tencent is secretly developing a “top-secret” WeChat AI agent capable of connecting millions of Mini Programs, with the potential in the future to replace the need for its 1.4 billion monthly active users to manually complete tasks such as food delivery and ride-hailing ✨
Related link:
谷歌大幅更新 AI 设计工具 Stitch：首创“氛围设计”理念，颠覆传统线框图流程
简报：
- 谷歌更新 UI 设计工具 Stitch，核心引入“氛围设计”理念，允许用户打破传统线框图限制，直接从抽象概念和情感开启设计。
- 新版重构了用户界面并加入原生 AI 无限画布，支持图像、文本和代码等上下文；AI 可通过提问或语音指令实时设计和更新页面。
- 系统新增设计智能体和智能体管理器，可分析项目生命周期、协同解决设计难题，并支持从任意 URL 一键提取现成的设计系统。
- 推出对智能体友好的 DESIGN.md 格式文件以提升跨平台兼容性，同时支持将静态设计一键转化为可预览流程的交互式原型。
- Stitch 目前在 Google Labs 中作为实验性应用运行，开发者可通过新增的 MCP 服务器和 SDK 将 Cursor 等外部 AI 工具接入项目。
相关链接：
请问是否还需要为您整理其他相关科技资讯的简报？
Google Unveils Major Update to AI Design Tool Stitch: Introducing the Pioneering “Vibe Design” Concept, Disrupting the Traditional Wireframe Workflow ✨
Brief:
- Google has updated its UI design tool Stitch, with the core addition of the “vibe design” concept, allowing users to break free from the constraints of traditional wireframes and start designing directly from abstract ideas and emotions.
- The new version features a rebuilt user interface and a native AI infinite canvas, supporting context such as images, text, and code; the AI can design and update pages in real time through questions or voice commands.
- The system adds design agents and an agent manager, capable of analyzing the project lifecycle, collaboratively solving design challenges, and supporting one-click extraction of existing design systems from any URL.
- It also introduces the agent-friendly
DESIGN.mdfile format to improve cross-platform compatibility, while supporting one-click conversion of static designs into interactive prototypes with previewable flows. - Stitch is currently available as an experimental application in Google Labs, and developers can connect external AI tools such as Cursor to their projects through the newly added MCP server and SDK. 🤖
Related Links:
Would you like me to also compile briefings on other related tech news? 📌
如果对你有用的话，可以打赏哦
打赏
本文作者:topwind
本文链接:
版权声明:本博客所有文章除特别声明外，均采用 BY-NC-SA 许可协议。转载请注明出处！
目录