![[b940349a-43bf-4bfd-8259-e2c38fde4ff7.mp3]]

OpenAI发布新一代ChatGPT智能体，开启AI自动化办公新时代

简报：

• 2025年7月18日，OpenAI正式发布新一代ChatGPT Agent，集成了Operator网页交互与Deep Research深度分析能力，实现从对话助手迈向自主任务执行者的跨越。
• ChatGPT Agent支持主动作业，包括浏览网页、自动购物、生成可编辑PPT、处理复杂的办公和个人任务，并可访问Gmail、GitHub等应用进行多样化办公自动化。
• 在行业多项基准测试中，Agent表现优越：如“Humanity's Last Exam”测试准确率41.6%，在金融建模、电子表格编辑及Web任务等场景均超越微软Co-pilot等竞品。
• 用户通过自然语言指令即可启用Agent，享受包括自动报告生成、会议安排、数据分析等自动化生产力服务，且执行敏感操作需用户授权，保证数据安全与隐私。
• 此功能现已向ChatGPT Pro、Plus、Team用户开放，企业与教育用户将于7月后陆续获取权限，欧盟及瑞士地区暂未上线。未来计划集成更多功能，有望影响办公行业新格局。

相关链接：

• OpenAI重磅发布ChatGPT Agent：会主动思考，浏览、购物、做PPT
• OpenAI新一代ChatGPT智能体发布，开启自动化办公新时代
• ChatGPT Agent功能与性能详解
• OpenAI挑戰微軟Office領域，ChatGPT成為智慧辦公新神器
• OpenAI打造辦公生態圈，掀AI驅動革命

OpenAI Launches Next-Gen ChatGPT Agent, Ushering in a New Era of AI-Powered Office Automation

Briefing:

• On July 18, 2025, OpenAI officially released its next-generation ChatGPT Agent, integrating Operator's web interaction capabilities with Deep Research's in-depth analysis, marking a leap from conversational assistant to autonomous task executor. 🚀
• ChatGPT Agent supports proactive operations, including web browsing, automated shopping, generating editable PPTs, handling complex office and personal tasks, and can access applications like Gmail and GitHub for diverse office automation.
• The Agent demonstrated superior performance in various industry benchmarks: achieving 41.6% accuracy in the "Humanity's Last Exam" test and outperforming competitors like Microsoft Co-pilot in scenarios such as financial modeling, spreadsheet editing, and web tasks. ✨
• Users can activate the Agent via natural language commands to enjoy automated productivity services, including automatic report generation, meeting scheduling, and data analysis. Sensitive operations require user authorization to ensure data security and privacy.
• This feature is now available to ChatGPT Pro, Plus, and Team users, with enterprise and educational users gradually gaining access after July. It is not yet available in EU and Switzerland regions. Future plans include integrating more functions, potentially reshaping the office industry landscape. 🤖

Related Links:

• OpenAI Unveils ChatGPT Agent: Proactive Thinking, Browsing, Shopping, PPT Creation
• OpenAI's Next-Gen ChatGPT Agent Released, Ushering in a New Era of Automated Office
• ChatGPT Agent Features and Performance Explained
• OpenAI Challenges Microsoft Office Domain, ChatGPT Becomes a New Smart Office Tool
• OpenAI Builds Office Ecosystem, Sparking an AI-Driven Revolution

工信部加快推动5G、工业互联网与AI深度融合应用

简报：

• 7月18日，工信部信息通信发展司司长谢存表示，将持续推进网络建设升级，提升5G和千兆光网覆盖，加强5G-A和万兆光网试点，构建全国一体化算力网络体系。
• 工信部将深化“5G+工业互联网”工程，推动工业互联网与人工智能协同赋能，实现规模化应用。
• 同时，加速5G-A产品创新、推进6G研发、优化电信政策，鼓励外资企业参与相关项目试点。

相关链接：

• 工信部：推动工业互联网与人工智能协同赋能

MIIT Accelerates Deep Integration and Application of 5G, Industrial Internet, and AI 🚀

Briefing:

• On July 18th, Xie Cun, Director of the Department of Information and Communication Development at MIIT, stated that they will continue to promote network construction upgrades, enhance 5G and gigabit optical network coverage, strengthen 5G-Advanced (5G-A) and 10-gigabit optical network pilot programs, and build a national integrated computing power network system. 📶
• MIIT will deepen the "5G+Industrial Internet" project, promoting the synergistic empowerment of Industrial Internet and Artificial Intelligence to achieve large-scale application. 🏭
• Concurrently, efforts will be made to accelerate 5G-A product innovation, advance 6G R&D, optimize telecom policies, and encourage foreign-funded enterprises to participate in relevant project pilots.

Related Links:

• MIIT: Promote Synergistic Empowerment of Industrial Internet and Artificial Intelligence

开源AI视频生成新突破：LTX-Video 13B实现30倍速高清渲染

简报：

• Lightricks公司近日发布了开源高清视频生成模型LTX-Video 13B，具备130亿参数，并采用创新的多尺度渲染技术，实现比同类产品快30倍的生成速度，可在消费级GPU上高效运行。
• 该模型支持1216×704分辨率、30帧/秒的实时生成，支持关键帧控制、风格/动作替换等多种高级功能，为创作者提供了丰富的细粒度操作选项。
• LTX-Video 13B免费开放下载，适用于初创企业和小型组织，并可通过GitHub和Hugging Face获取与定制，进一步推动AI视频创作的普及和创新。

相关链接：

• LTX-Video 13B详细介绍与新闻报道
• Lightricks发布13B参数LTX-Video，显著提升视频AI生成效率
• Lightricks shakes up AI video creation with powerful open-source model
• abc27商业新闻稿

Open-Source AI Video Generation Breakthrough: LTX-Video 13B Achieves 30x Faster HD Rendering

Briefing:

• Lightricks recently released LTX-Video 13B, an open-source high-definition video generation model with 13 billion parameters. It leverages innovative multi-scale rendering technology to achieve generation speeds 30 times faster than comparable models, operating efficiently on consumer-grade GPUs. 🚀
• This model supports real-time generation at 1216×704 resolution and 30 frames per second. It offers advanced features such as keyframe control and style/motion replacement, providing creators with extensive fine-grained operational options. 🎬
• LTX-Video 13B is freely available for download, suitable for startups and small organizations. It can be accessed and customized via GitHub and Hugging Face, further promoting the widespread adoption and innovation in AI video creation. ✨

Related Links:

• LTX-Video 13B detailed introduction and news report
• Lightricks just made AI video generation 30x faster – and you won’t need a $10,000 GPU
• Lightricks shakes up AI video creation with powerful open-source model
• abc27 Business Press Release

NVIDIA发布新一代ASR模型Canary-Qwen-2.5B，刷新语音识别准确率纪录

简报：

• NVIDIA正式推出商用级超高速自动语音识别（ASR）模型Canary-Qwen-2.5B，以5.63%的词错误率（WER）创下Hugging Face OpenASR排行榜历史新低。
• Canary-Qwen-2.5B集成ASR与大型语言模型功能，实现音频转录、摘要、问答等多模态任务的统一处理，支持企业级多场景应用，具备开源及商业许可（CC-BY）。
• 新模型采用创新混合架构，具备25亿参数、基于234,000小时英语语音数据训练，并优化兼容多款NVIDIA GPU，适合数据中心及边缘部署，推动语音AI行业发展。

相关链接：

• NVIDIA Canary-Qwen-2.5B Hugging Face页面
• 新闻报道1
• 新闻报道2

NVIDIA Launches New-Generation ASR Model Canary-Qwen-2.5B, Setting New Accuracy Record for Speech Recognition

Brief:

• NVIDIA has officially introduced its commercial-grade ultra-fast Automatic Speech Recognition (ASR) model, Canary-Qwen-2.5B, which sets a new historical low Word Error Rate (WER) of 5.63% on the Hugging Face OpenASR Leaderboard. 🚀
• Canary-Qwen-2.5B integrates ASR with Large Language Model capabilities, enabling unified processing of multi-modal tasks such as audio transcription, summarization, and Q&A. It supports enterprise-grade multi-scenario applications and is available under open-source and commercial licenses (CC-BY). 💡
• The new model features an innovative hybrid architecture with 2.5 billion parameters, trained on 234,000 hours of English speech data. It is optimized for compatibility with multiple NVIDIA GPUs, making it suitable for data center and edge deployments, thereby advancing the speech AI industry. 📈

Related Links:

• NVIDIA Canary-Qwen-2.5B Hugging Face Page
• News Report 1
• News Report 2