![[aa76539e-9158-40b6-ab35-102bbe246cbb.mp3]]

腾讯开源工业级3D生成大模型混元3D 2.1，提升细节与材质表现

简报：

• 在CVPR2025大会上，腾讯宣布对外开源混元3D 2.1大模型，为业内首个全链路开源的工业级3D生成模型，支持模型权重、训练代码和数据流程。
• 新版本优化了几何生成质量、提升PBR材质真实感和细节表现，支持高质量渲染与多种复杂材质，广泛适用于游戏、电影、线上商城和数字孪生等领域。
• 混元3D 2.1降低了开发门槛，兼容消费级显卡，便于企业与个人开发者部署与定制，相关资源在Github及官网开放下载。
• 该模型在知名开源社区Hugging Face下载量超180万，满足学术及工业场景的3D资产需求。

相关链接：

• 腾讯宣布混元3D 2.1大模型对外开源
• 开源与体验官网
• 在线体验地址

Tencent Open-Sources Industrial-Grade 3D Generative AI Model Hunyuan3D 2.1, Enhancing Detail and Material Representation

Brief:

• At the CVPR 2025 conference, Tencent announced the open-sourcing of the Hunyuan3D 2.1 large model, making it the industry's first full-stack open-source industrial-grade 3D generative model, supporting model weights, training code, and data pipelines. 🚀
• The new version optimizes geometric generation quality, enhances PBR (Physically Based Rendering) material realism and detail representation, and supports high-quality rendering with various complex materials, making it widely applicable in fields such as gaming, film, online retail, and digital twins. ✨
• Hunyuan3D 2.1 lowers the development barrier and is compatible with consumer-grade graphics cards, facilitating deployment and customization for businesses and individual developers. Relevant resources are available for download on GitHub and the official website. 🛠️
• The model has garnered over 1.8 million downloads on the renowned open-source community Hugging Face, meeting the demand for 3D assets in both academic and industrial scenarios.

Related Links:

• Tencent Announces Open Source of Hunyuan3D 2.1 Large Model
• Open Source and Experience Official Website
• Online Experience Address

国产AI模型高考数学成绩排名公布，夸克表现最佳

简报：

• 国产AI参与高考数学能力测试，夸克AI在解题准确率和答题速度上均领先于豆包和元宝AI。
• 本次评测涵盖2024年高考数学新题，以全面考察各AI模型在数学理解和运算能力上的表现。
• 结果显示，夸克AI解答准确性和效率高于其它两款国产AI，凸显其算法优化和持续学习能力。

相关链接：

• 国产AI高考数学实力比拼：夸克领先豆包、元宝

Domestic AI Model Gaokao Math Scores Ranking Announced, Quark Performs Best! 🥇

Brief:

• Domestic AI models participated in the Gaokao Math ability test, with Quark AI leading both in problem-solving accuracy and answering speed, outperforming Doubao and Yuanbao AI. 🚀
• This evaluation covered new questions from the 2024 Gaokao Math exam to comprehensively assess each AI model's performance in mathematical understanding and computational abilities.
• The results indicate that Quark AI's problem-solving accuracy and efficiency were higher than the other two domestic AI models, highlighting its optimized algorithms and continuous learning capabilities. ✨

Related Link:

• Domestic AI Gaokao Math Strength Comparison: Quark Leads Doubao, Yuanbao

黄仁勋认为华为可满足中国AI需求并解读任正非“集群补代差”观点

简报：

• 英伟达CEO黄仁勋近日回应任正非最新讲话，指出虽然英伟达技术领先一代，但人工智能可以通过并行运算和更多芯片实现弥补。
• 黄仁勋强调，中国技术对本土市场已“足够好”，华为有能力覆盖中国以及部分全球市场的AI需求。
• 他表示，如果美国选择不参与中国市场，华为等中国公司仍能满足国内人工智能发展需要，不必过度担心单颗芯片落后带来的影响。
• 黄仁勋还指出，美国若放弃中国市场可能失去全球AI生态主导地位，呼吁关注长期战略影响。

相关链接：

• 36氪新闻原文
• SETN报道1
• SETN报道2

Jensen Huang Believes Huawei Can Meet China's AI Needs and Interprets Ren Zhengfei's 'Cluster Compensation for Generational Gaps' View

Briefing:

• NVIDIA CEO Jensen Huang recently responded to Ren Zhengfei's latest speech, stating that while NVIDIA's technology is a generation ahead, AI performance can be compensated through parallel computing and more chips. 💡
• Huang emphasized that Chinese technology is "good enough" for the domestic market, and Huawei is capable of covering AI demand in China and even parts of the global market.
• He stated that if the US chooses not to participate in the Chinese market, Chinese companies like Huawei can still meet the domestic AI development needs, and there's no need to overly worry about the impact of a single chip being behind. 🇨🇳
• Huang also pointed out that if the US abandons the Chinese market, it might lose its dominant position in the global AI ecosystem, calling for attention to long-term strategic implications. 🌍

Related Links:

• 36Kr News Original
• SETN Report 1
• SETN Report 2

ChatGPT Projects功能大升级：引入深度研究与语音模式，打造多模态AI工作中心

简报：

• OpenAI在“12Days of OpenAI”活动中推出ChatGPT Projects重大功能更新，将其从基础组织工具升级为集成搜索、语音、记忆优化等多项AI能力的智能工作空间。
• 新增“深度研究”功能，支持结合项目上下文与互联网搜索，实现更精准的信息检索和内容整合，适用学术、商业分析等高复杂需求场景。
• 集成语音模式，用户可用语音直接与项目交互，支持移动端办公和实时团队协作，提升交互便捷性。
• 项目记忆能力优化，能智能引用历史对话与数据，辅助持续任务管理与建议。
• 移动端现可上传文件、切换模型及进行实时视频和屏幕共享，拓展移动使用场景。
• 项目升级后适用个人、企业及教育等多样场景，并陆续上线企业级连接器，实现与内部数据系统集成。
• OpenAI强调数据安全与合规，功能全球推广正在逐步展开。

相关链接：

• OpenAI Projects重大更新详解：深度研究、语音模式与多模态协作
• OpenAI发布ChatGPT Projects，支持智能协作与文件管理
• ChatGPT Projects深度评测：多场景应用与功能亮点

ChatGPT Projects Undergoes Major Upgrade: Deep Research & Voice Mode Introduced, Building a Multimodal AI Work Hub 🚀

Brief:

• OpenAI has unveiled a major feature update for ChatGPT Projects during its "12 Days of OpenAI" event, transforming it from a basic organizational tool into an intelligent workspace integrating AI capabilities such as search, voice, and memory optimization. 🧠
• A new "Deep Research" feature has been added, supporting more precise information retrieval and content integration by combining project context with internet search, suitable for high-complexity scenarios like academic research and business analysis.
• Integrated voice mode allows users to interact directly with projects via voice, supporting mobile work and real-time team collaboration, enhancing interaction convenience.
• Project memory capabilities have been optimized, intelligently referencing historical conversations and data to assist with continuous task management and suggestions.
• Mobile versions now support file uploads, model switching, and real-time video and screen sharing, expanding mobile usage scenarios.
• Post-upgrade, Projects is applicable across diverse scenarios including personal, enterprise, and education, with enterprise-grade connectors being rolled out to enable integration with internal data systems.
• OpenAI emphasizes data security and compliance, with global rollout of the features gradually underway. 🌐

Related Links:

• OpenAI Projects重大更新详解：深度研究、语音模式与多模态协作
• OpenAI发布ChatGPT Projects，支持智能协作与文件管理
• ChatGPT Projects深度评测：多场景应用与功能亮点

AMD与OpenAI携手发布新一代AI芯片，推理性能大幅提升

简报：

• 在2025全球AI发展大会上，AMD与OpenAI联合发布了Instinct MI350和MI400系列AI芯片，重点提升AI推理及计算性能。
• MI350系列GPU基于全新CDNA4架构，AI推理性能相比上一代提升35倍，配备288GB HBM3E内存和高达8TB/s带宽。
• 新芯片支持多种冷却部署方式，并通过开源AI加速平台ROCm7, 带来超过3.5倍推理性能提升。
• MI400系列作为下一代旗舰，优化AI低精度训练，支持72块GPU无缝互联，满足大规模AI训练需求。
• 多家企业如Oracle、微软、Meta和xAI已与AMD展开合作，Oracle将在其云服务中率先部署MI355X解决方案。

相关链接：

• AIBase新闻报道
• IT之家相关新闻

AMD and OpenAI Collaborate to Unveil Next-Gen AI Chips, Boosting Inference Performance Significantly 🚀

Briefing:

• At the 2025 Global AI Development Conference, AMD and OpenAI jointly announced the Instinct MI350 and MI400 series AI chips, with a focus on enhancing AI inference and computational performance.
• The MI350 series GPUs, based on the new CDNA4 architecture, deliver a 35x improvement in AI inference performance compared to the previous generation, featuring 288GB HBM3E memory and up to 8TB/s bandwidth.
• The new chips support various cooling deployment methods and offer over 3.5x inference performance improvement through the open-source AI acceleration platform ROCm7.
• The MI400 series, as the next-generation flagship, optimizes AI low-precision training and supports seamless interconnection of 72 GPUs, meeting the demands of large-scale AI training.
• Several companies, including Oracle, Microsoft, Meta, and xAI, have initiated collaborations with AMD. Oracle will be the first to deploy the MI355X solution in its cloud services. ✨🧠

Related Links:

• AIBase News Report
• ITHome Related News

MiniMax Agent重大升级，推出多模态AI新功能强化高效创作

简报：

• MiniMax官方宣布其AI助手产品MiniMax Agent迎来重大升级并已正式上线（https://agent.minimax.io）。
• 新版本集成智能图像搜索、更稳定的图像生成、反射模式（提升长期和多步骤任务处理能力）、Python绘图支持（兼容中文、日文、韩文字体）等功能。
• 用户可直接将成果导出为PDF、DOCX和PPTX格式，方便文档分享与专业展示。
• MiniMax Agent基于多模态技术及多代理协作协议，支持文本、图像、语音等多种内容处理，为个人及企业用户提供提升创作与生产力的一站式解决方案。

相关链接：

• MiniMax Agent正式上线，多模态能力强化AI生产力
• MiniMax Agent重大升级，图像生成和多语言支持增强

MiniMax Agent Undergoes Major Upgrade, Introduces Multimodal AI Features to Enhance Efficient Creation 🚀

Briefing:

• MiniMax has officially announced a significant upgrade and launch of its AI assistant product, MiniMax Agent (https://agent.minimax.io).
• The new version integrates features such as intelligent image search, more stable image generation, Reflection Mode (enhancing long-term and multi-step task processing capabilities), and Python plotting support (compatible with Chinese, Japanese, and Korean fonts).
• Users can directly export results in PDF, DOCX, and PPTX formats, facilitating easy document sharing and professional presentations. 💼
• Built on multimodal technology and a multi-agent collaboration protocol, MiniMax Agent supports various content types including text, images, and voice, offering a one-stop solution for individuals and enterprises to boost creativity and productivity. ✨

Related Links:

• MiniMax Agent Officially Launched, Multimodal Capabilities Boost AI Productivity
• MiniMax Agent Undergoes Major Upgrade, Enhancing Image Generation and Multilingual Support