单词学习：![[7d5292d7-30c3-4bdd-afa4-15748bf010bf.mp3]]

新闻播报：![[6125d173-50d2-4ac1-b42c-14aa4ea5d74e.mp3]]

腾讯混元3D 3.0模型发布，建模精度提升三倍且免费使用

简报：

• 在2025腾讯全球数字生态大会上，腾讯正式发布混元3D 3.0模型，其首创的3D-DiT分级雕刻技术使建模精度较前代提升3倍，几何分辨率达1536³。
• 该模型针对人物生成进行专项优化，显著提升了面部、五官轮廓的精细度和体态的自然流畅度，并优化了纹理贴合的真实感。
• 混元3D 3.0模型现已集成于混元3D AI创作引擎，向用户免费开放，并同步上线腾讯云API。
• 配套的专业级3D AI工作台“混元3D Studio”已开启邀请制内测，同时腾讯计划近期开源支持多条件控制的混元3D Omni模型。

相关链接：

• 腾讯重磅发布混元3D 3.0模型 建模精度提升3倍
• 腾讯混元 3D 3.0 AI 模型发布：建模精度提升 3 倍，免费开放使用

Tencent Hunyuan 3D 3.0 Model Launched, Modeling Accuracy Tripled and Free to Use

Brief:

• At the 2025 Tencent Global Digital Ecosystem Summit, Tencent officially unveiled the Hunyuan 3D 3.0 model. Its innovative 3D-DiT hierarchical sculpting technology boosts modeling accuracy by 3 times compared to its predecessor, achieving a geometric resolution of 1536³. ✨
• The model features specialized optimization for human character generation, significantly enhancing the delicacy of facial and feature contours, the natural fluidity of body posture, and optimizing the realism of texture mapping. 🧑‍🎨
• The Hunyuan 3D 3.0 model is now integrated into the Hunyuan 3D AI Creation Engine, made available for free to users, and simultaneously launched on Tencent Cloud API. 🚀
• The accompanying professional-grade 3D AI workbench, "Hunyuan 3D Studio," has initiated an invitation-only internal beta. Additionally, Tencent plans to open-source the Hunyuan 3D Omni model, which supports multi-conditional control, in the near future.

Related Links:

• Tencent Releases Hunyuan 3D 3.0 Model, Modeling Accuracy Increased by 3 Times
• Tencent Hunyuan 3D 3.0 AI Model Released: Modeling Accuracy Tripled, Available for Free

腾讯发布AI繁荣计划，赋能超300家企业开发智能体

简报：

• 9月16日，腾讯在全球数字生态大会上正式发布“AI应用繁荣计划”，已吸引超过300家企业报名参与。
• 该计划包含“AI共创营”和“AI百校行”两大模块，旨在孵化金融、法律、制造等垂直场景中的智能体与大模型应用。
• 计划将围绕腾讯混元大模型、腾讯云智能体开发平台、腾讯元器三大核心产品，为合作伙伴提供技术、云资源和培训支持。
• 已有标杆案例落地，如得理科技的AI合同审核准确率达95%，创想三维实现5分钟内照片生成3D手办模型。

相关链接：

• 腾讯启动AI应用繁荣计划，超300家企业抢滩智能体新赛道

Tencent Launches AI Prosperity Plan, Empowering Over 300 Enterprises to Develop Intelligent Agents 🚀

Brief:

• On September 16th, Tencent officially launched the "AI Application Prosperity Plan" at its Global Digital Ecosystem Summit, attracting over 300 enterprises to sign up.
• The plan comprises two main modules: "AI Co-creation Camp" and "AI Hundred Schools Initiative," aiming to incubate intelligent agents and large model applications in vertical scenarios such as finance, law, and manufacturing. 💡
• The initiative will leverage three core products – Tencent Hunyuan Large Model, Tencent Cloud Intelligent Agent Development Platform, and Tencent Yuanshi – to provide partners with technical, cloud resource, and training support.
• Benchmark cases have already emerged, such as Delir Technology achieving a 95% accuracy rate in AI contract review, and Creality 3D generating 3D figurine models from photos within 5 minutes. ✨

Related Links:

• Tencent Launches AI Application Prosperity Plan, Over 300 Enterprises Vie for New Intelligent Agent Track

逻辑智能推出全球首个开源语音大模型框架LLaSO

简报： 北京深度逻辑智能科技有限公司于近日发布了全球首个完全开放、端到端的开源语音语言模型框架LLaSO，旨在通过开放访问、修改和使用权限，降低技术门槛并推动语音AI技术的普及与共享。

相关链接：

• 逻辑智能发布全球首个开源语音大模型框架LLaSO

Deep Logic AI Launches World's First Open-Source Large Voice Model Framework LLaSO

Brief: Beijing Deep Logic AI Technology Co., Ltd. recently released LLaSO, the world's first fully open, end-to-end open-source speech language model framework. 🎉 It aims to lower technical barriers and promote the popularization and sharing of speech AI technology by providing open access, modification, and usage rights. 🚀🌐

Related links:

• 逻辑智能发布全球首个开源语音大模型框架LLaSO

上海AI Lab发布Lumina-DiMOO，以创新架构提升多模态处理效率

简报：

• 上海人工智能实验室联合多所高校推出新一代多模态模型Lumina-DiMOO。
• 该模型采用创新的“全离散扩散架构”，旨在突破传统模型在处理文本与图像等不同模态数据时的局限。
• Lumina-DiMOO通过将文本、图像等数据映射到共享的“语义空间”，并利用对比学习技术，实现了多模态数据的有效整合、理解与生成。
• 模型在图像生成任务中兼顾了速度与质量，少量步骤即可获得高质量结果，并在文本到图像生成、图像理解与分析等场景表现出色。

相关链接：

• 上海AI Lab推出Lumina-DiMOO，开创多模态生成与理解新局面
• 项目地址

Shanghai AI Lab Unveils Lumina-DiMOO, Boosting Multimodal Processing Efficiency with Innovative Architecture

Brief:

• Shanghai AI Lab, in collaboration with several universities, has launched Lumina-DiMOO, a new generation multimodal model.
• The model adopts an innovative "fully discrete diffusion architecture" 🧠, aiming to overcome the limitations of traditional models in processing diverse modalities like text and images.
• Lumina-DiMOO effectively integrates, understands, and generates multimodal data by mapping text, images, and other data into a shared "semantic space" and leveraging contrastive learning techniques.
• The model balances speed and quality in image generation tasks, achieving high-quality results with fewer steps. It also demonstrates excellent performance in scenarios such as text-to-image generation, image understanding, and analysis. ✨🚀

Related Links:

• Shanghai AI Lab Launches Lumina-DiMOO, Opening New Horizons for Multimodal Generation and Understanding
• Project Repository

QuestMobile发布AI应用榜单：豆包月活登顶，AQ增速第一

简报：

• 根据QuestMobile最新报告，豆包以1.57亿月活跃用户规模位列中国AI原生应用榜首，超越DeepSeek。
• 腾讯元宝、Kimi和即梦AI位列其后，其中元宝月活环比增速达22.4%。
• 蚂蚁集团旗下AI健康应用AQ首次入榜，并以超60.1%的环比增速位列全行业第一。
• 报告显示，AI原生应用中专业顾问赛道潜力巨大，该领域原生App环比增速达56.5%。

相关链接：

• QuestMobile发布中国AI原生应用Top10：豆包、元宝、AQ入选
• QuestMobile 数据：豆包月活超越DeepSeek，荣登中国原生 AI APP 榜首

QuestMobile Unveils AI App Rankings: Doubao Tops MAU, AQ Leads Growth 🚀

Brief:

• According to QuestMobile's latest report, Doubao leads China's AI-native applications with 157 million monthly active users (MAU), surpassing DeepSeek.
• Tencent Yuanbao, Kimi, and Jimeng AI follow, with Yuanbao's MAU experiencing a month-on-month growth of 22.4%.
• Ant Group's AI health application, AQ, makes its debut on the list, boasting the industry's highest month-on-month growth rate, exceeding 60.1%! 📈
• The report highlights immense potential in the professional consultant track within AI-native applications, with native apps in this sector achieving a remarkable 56.5% month-on-month growth. 💡

Related Links:

• QuestMobile Releases China's Top 10 AI-Native Apps: Doubao, Yuanbao, AQ Included
• QuestMobile Data: Doubao Surpasses DeepSeek in MAU, Crowned China's Top Native AI App

新版人工智能安全治理框架发布，优化风险分类与治理措施

简报：

• 《人工智能安全治理框架》2.0版于2025年9月15日在国家网络安全宣传周主论坛上正式发布。
• 该框架由国家互联网应急中心牵头，在国家网信办指导下制定，旨在应对人工智能技术和应用迅速发展带来的新机遇与挑战，是对2024年1.0版的更新。
• 2.0版结合人工智能技术发展和应用实践，完善优化了风险分类，研究探索风险分级，并动态调整更新了防范治理措施。

相关链接：

• 《人工智能安全治理框架》2.0 版发布

Updated AI Safety Governance Framework Released, Optimizing Risk Classification and Governance Measures 🛡️

Brief:

• The AI Safety Governance Framework 2.0 was officially released on September 15, 2025, at the main forum of the National Cybersecurity Awareness Week. 🚀
• The framework was led by the National Computer Network Emergency Response Technical Team/Coordination Center of China (CNCERT/CC) and formulated under the guidance of the Cyberspace Administration of China (CAC). It aims to address the new opportunities and challenges presented by the rapid development of AI technology and applications, serving as an update to the 1.0 version from 2024.
• Version 2.0, incorporating the evolution of AI technology and application practices, has refined and optimized risk classification, explored risk grading, and dynamically adjusted and updated prevention and governance measures. 💡

Related links:

• AI Safety Governance Framework 2.0 Released

挑战AI黑箱：瑞士推出全开源模型Apertus

简报：

• 瑞士联邦理工学院洛桑分校（EPFL）、苏黎世联邦理工学院（ETH Zurich）及瑞士国家超级计算中心（CSCS）联合发布了名为Apertus的大规模开源语言模型。
• Apertus在拉丁语中意为“开放”，其目标是通过前所未有的透明度，挑战当前主流AI模型的“黑箱”现状。
• 该项目全面公开了模型权重、架构、训练代码、数据配方以及完整的训练过程文档，旨在促进全球范围内的合作与创新。

相关链接：

• 瑞士Apertus开源模型挑战AI“黑箱”，全面公开训练细节

Challenging the AI Black Box: Switzerland Launches Fully Open-Source Model Apertus

Brief:

• EPFL (École Polytechnique Fédérale de Lausanne), ETH Zurich (Eidgenössische Technische Hochschule Zürich), and CSCS (Swiss National Supercomputing Centre) have jointly released a large-scale open-source language model named Apertus.
• Apertus, meaning 'open' in Latin, aims to challenge the 'black box' status quo of current mainstream AI models through unprecedented transparency. 🔓
• The project has fully disclosed model weights, architecture, training code, data recipes, and comprehensive documentation of the training process, aiming to foster global collaboration and innovation. 🌍💡

Related links:

• Swiss Apertus Open-Source Model Challenges AI 'Black Box', Fully Discloses Training Details