![[1894cbaf-f780-4131-b449-3dea536dfbe0.mp3]]

华为云盘古5.5大模型发布，统一三元组编码提升跨行业预测能力

简报：

• 在华为开发者大会2025上，华为云正式发布盘古大模型5.5，采用首创Triplet Transformer（三元组编码）统一预训练架构，实现对表格、时序和图像等多类型数据的高效统一处理，大幅提升预测精度。
• 新一代盘古大模型实现自然语言处理、视觉、多模态、预测、科学计算五大基础模型全面升级，其中五款具备深度思考能力的行业自然语言大模型（医学、金融、政务、工业、汽车）即将上线。
• 盘古5.5通过提升模型泛化和推理能力，助力企业在多行业场景下实现高效智能化转型和数字升级。

相关链接：

• 华为云发布盘古5.5：三元组编码架构引领预测大模型新潮流
• 华为发布盘古大模型5.5：五大基础模型升级

Huawei Cloud Unveils Pangu Model 5.5: Unified Triplet Encoding Boosts Cross-Industry Prediction Capabilities

Briefing:

• At Huawei Developer Conference 2025, Huawei Cloud officially unveiled Pangu Model 5.5. It adopts the groundbreaking Triplet Transformer (triplet encoding) unified pre-training architecture, achieving efficient and unified processing of multi-type data such as tables, time series, and images, significantly enhancing prediction accuracy. 🚀
• The new generation Pangu Model achieves comprehensive upgrades across five foundational models: natural language processing, vision, multimodal, prediction, and scientific computing. Five industry-specific natural language large models with deep thinking capabilities (medical, financial, government affairs, industrial, and automotive) are set to be launched soon. 💡
• Pangu 5.5, by enhancing its model generalization and inference capabilities, assists enterprises in achieving efficient intelligent transformation and digital upgrading across various industry scenarios. ✨

Related Links:

• Huawei Cloud Releases Pangu 5.5: Triplet Encoding Architecture Leads New Trend in Predictive Large Models
• Huawei Releases Pangu Model 5.5: Five Foundational Models Upgraded

Kimi-Researcher深度研究Agent开启灰度测试，实现高效自主信息筛选与深度报告生成

简报：

• 月之暗面旗下Kimi智能助手正式推出首个基于端到端强化学习的Agent产品——Kimi-Researcher，并开启小范围灰度测试。
• Kimi-Researcher展现出自主规划、澄清问题、关键词搜索与高质量信息筛选的能力，平均每项任务可进行23步推理、规划74个关键词、筛选206个网址，最终保留信息质量最高的3.2%内容。
• 该Agent可自动调用浏览器和代码工具、处理原始数据，并生成信息详实且可溯源的深度研究报告，支持交互、可视化与便捷分享。
• 在AI高难度benchmark“Humanity’s Last Exam”测试中，Kimi-Researcher以Pass@1 26.9%、Pass@4 40.17%的成绩刷新多项记录，表现优异。
• 内测资格可通过kimi.com申请。

相关链接：

• 月之暗面Kimi-Researcher灰度测试启动

Kimi-Researcher Deep Research Agent Initiates Gray Test, Achieving Efficient Autonomous Information Filtering and In-depth Report Generation 🔍

Brief:

• Moonshot AI's Kimi intelligent assistant officially launched its first end-to-end reinforcement learning-based Agent product, Kimi-Researcher, initiating a small-scale gray test. ✨
• Kimi-Researcher demonstrates capabilities in autonomous planning, problem clarification, keyword search, and high-quality information filtering. On average, for each task, it performs 23 inference steps, plans 74 keywords, screens 206 URLs, ultimately retaining the top 3.2% of the highest quality information.
• This Agent can automatically call browsers and coding tools, process raw data, and generate detailed and traceable in-depth research reports, supporting interactivity, visualization, and convenient sharing.
• In the challenging AI benchmark "Humanity’s Last Exam" test, Kimi-Researcher set new records with Pass@1 26.9% and Pass@4 40.17%, demonstrating excellent performance. 🚀
• Internal testing qualification can be applied for via kimi.com.

Related Links:

• Moonshot AI Kimi-Researcher Gray Test Launched

MiniMax推视频Agent工具，支持文字指令和人脸一致性生成高清视频

简报：

• MiniMax近日发布视频Agent工具，用户可通过一句文本指令生成最高清720p、25帧/秒、最长6秒的高清视频，未来将扩展至10秒。
• 工具支持上传人脸图片实现视频中人物身份的精准一致性，便于虚拟主播、品牌广告等个性化视频创作。
• 该工具依托MiniMax多模态AI技术，并通过开放API接口，方便企业和开发者集成使用；自发布以来，Hailuo平台已生成超37亿个视频。
• 视频Agent工具在业内具有视频创作门槛低、人脸一致性强、生成成本较低等优势，但目前视频时长和多语言支持仍待优化。

相关链接：

• MiniMax重磅发布视频Agent工具
• MiniMax推出Hailuo Video Agent开放Beta版
• Hailuo 02用户基准测试超越Google Veo 3

MiniMax Launches Video Agent Tool: Text-to-Video Generation with Consistent Facial Identity

Briefing:

• MiniMax recently unveiled its Video Agent tool, enabling users to generate high-definition videos up to 720p, 25 frames/second, and a maximum of 6 seconds, from a single text prompt. Future plans include extending video duration to 10 seconds. 🎥
• The tool supports uploading face images to ensure precise identity consistency for characters within videos, facilitating personalized video creation for virtual anchors, brand advertisements, and more.
• This tool is powered by MiniMax's multimodal AI technology and offers an open API interface for easy integration by businesses and developers. Since its launch, the Hailuo platform has generated over 3.7 billion videos. ✨
• The Video Agent tool offers industry advantages such as a low barrier to video creation, strong facial consistency, and relatively low generation costs. However, current video duration and multilingual support are still areas for optimization. 💡

Related Links:

• MiniMax Unveils Its Powerful Video Agent Tool
• MiniMax Launches Hailuo Video Agent Open Beta
• Hailuo 02 User Benchmark Exceeds Google Veo 3

B站上线AI智能体提升商业平台效率

简报：

• B站近期接入通义千问Qwen 3、Qwen-VL等大模型，推出数据洞察智能体InsightAgent，为其商业平台"花火"与"必达"显著提升了效率。
• 在618电商大促期间，InsightAgent令商单成交和数据洞察效率大幅提升，其中"花火"平台AI选人功能帮助商单成交效率提升超5倍。
• "必达"平台应用InsightAgent，实现了智能报告生成和行业趋势分析，显著缩短品牌投放决策时间。

相关链接：

• B站接入通义千问Qwen 3等模型 推出数据洞察智能体InsightAgent

Bilibili Launches AI Agents to Boost Commercial Platform Efficiency

Brief:

• Bilibili recently integrated large models such as Tongyi Qianwen Qwen 3 and Qwen-VL, launching the data insight AI agent InsightAgent. This has significantly boosted the efficiency of its commercial platforms, "Huahuo" and "Bida"!
• During the 618 e-commerce grand promotion, InsightAgent led to a substantial improvement in business order conversion and data insight efficiency. Notably, the AI creator selection feature on the "Huahuo" platform helped boost business order conversion efficiency by over 5 times. 🚀
• The "Bida" platform, utilizing InsightAgent, has achieved intelligent report generation and industry trend analysis, significantly reducing brand advertising decision-making time. 💡

Related Link:

• Bilibili Integrates Tongyi Qianwen Qwen 3 and Other Models, Launches Data Insight Agent InsightAgent

ChatGPT上线macOS专属笔记工具，支持会议自动转录与智能总结

简报：

• OpenAI 推出专为会议记录、头脑风暴和个人笔记管理设计的 ChatGPT 新笔记工具，现已在 macOS 桌面应用程序中上线。
• 用户可通过“隐形”录音模式自动转录会议音频，并实时生成结构化会议笔记，支持按需提取关键讨论内容和行动项，无需额外机器人介入，保证信息私密性。
• 新功能已于2025年6月4日率先向 Team 用户提供，6月18日起扩展至 Pro、Enterprise 和 Edu 用户，暂时仅限 macOS 平台。
• 此工具可结合 ChatGPT 记忆功能，实现更个性化的笔记整理，市场反馈显示其智能总结功能被广泛认可，但与同类产品在协作与多平台支持方面仍有差距。

相关链接：

• AIbase新闻
• 三立新闻网
• IT之家报道

ChatGPT Launches macOS-Exclusive Note-Taking Tool, Featuring Automatic Meeting Transcription and Smart Summarization

Briefing:

• OpenAI has launched a new ChatGPT note-taking tool specifically designed for meeting notes, brainstorming, and personal note management, now available within its macOS desktop application. 📝
• Users can automatically transcribe meeting audio via an "invisible" recording mode, generating structured meeting notes in real-time. It supports on-demand extraction of key discussions and action items without requiring additional bot intervention, ensuring information privacy. 🤫
• The new feature was first made available to Team users on June 4, 2025, and will be extended to Pro, Enterprise, and Edu users starting June 18. It is currently limited to the macOS platform. 📅
• This tool can be combined with ChatGPT's memory function for more personalized note organization. Market feedback indicates widespread recognition for its smart summarization capabilities, though it still lags behind similar products in terms of collaboration and multi-platform support.

Related Links:

• AIbase News
• Sanli News Network
• IT Home Report