单词学习：![[2e7872e3-1069-43a9-a99a-ac1653906988.mp3]]

新闻播报：![[fe63a36b-cc20-47ed-80db-d994a363d1d6.mp3]]

MiniMax开源M2模型：成本仅为Claude Sonnet的8%

简报：

• 2025年10月27日，中国AI初创公司MiniMax宣布开源大语言模型MiniMax M2，专为代理工作流和端到端编码任务设计
• M2采用混合专家（MoE）架构，总参数量达2300亿，推理时仅激活100亿参数，支持204,800令牌的上下文窗口
• 模型每令牌成本仅为Anthropic Claude Sonnet的8%，速度提升约两倍，在编码能力、代理性能方面表现优异
• 根据Artificial Analysis基准测试，M2在数学、科学、编码和代理工具使用等综合智能指标中位列全球开源模型之首
• M2采用Apache 2.0许可证开源，模型权重已上架Hugging Face，通过MiniMax平台和API提供全球限时免费访问

相关链接：

• MiniMax开源M2模型详情报道
• MiniMax M2模型Hugging Face下载地址
• MiniMax平台文档

MiniMax Open-Sources M2 Model: 8% the Cost of Claude Sonnet

Briefing:

• On October 27, 2025, Chinese AI startup MiniMax announced the open-sourcing of its large language model, MiniMax M2, designed specifically for agentic workflows and end-to-end coding tasks. 🚀
• M2 utilizes a Mixture-of-Experts (MoE) architecture with a total of 230 billion parameters, activating only 10 billion parameters during inference, and supports a 204,800-token context window.
• The model's per-token cost is only 8% that of Anthropic's Claude Sonnet, and it boasts approximately twice the inference speed, demonstrating excellent performance in coding capabilities and agentic tasks. 💰
• According to Artificial Analysis benchmarks, M2 ranks first among global open-source models in comprehensive intelligence metrics including mathematics, science, coding, and agent tool usage.
• M2 is open-sourced under the Apache 2.0 license, with model weights now available on Hugging Face. Global limited-time free access is also provided via the MiniMax platform and API. ✨

Related Links:

• Detailed Report on MiniMax M2 Open-Source Model
• MiniMax M2 Model Download on Hugging Face
• MiniMax Platform Documentation

OpenAI推出Company Knowledge，让ChatGPT化身企业知识中枢

简报：

• OpenAI于2025年10月26日面向Business、Enterprise及Edu订阅用户正式推出“Company Knowledge”（公司知识）新功能。
• 该功能允许ChatGPT直接连接企业授权的Slack、Google Drive、Salesforce等12个主流协作平台，实现跨系统数据查询。
• 用户可通过自然语言提问，AI能自动整合多源信息，生成附带引用来源的精准回答，旨在打破企业“知识孤岛”，提升决策效率。
• 功能由专为多源检索优化的GPT-5增强版模型驱动，并承诺所有企业数据仅用于即时查询，绝不用于模型训练，以保障商业机密安全。

相关链接：

• ChatGPT“企业大脑”正式上线！OpenAI推Company Knowledge功能，一键打通Slack、Drive等12大平台，决策效率飙升

OpenAI Launches Company Knowledge, Turning ChatGPT into an Enterprise Knowledge Hub

Brief:

• On October 26, 2025, OpenAI officially launched its new "Company Knowledge" feature for Business, Enterprise, and Edu subscribers. 🎉
• This feature allows ChatGPT to directly connect to 12 major collaborative platforms authorized by enterprises, including Slack, Google Drive, and Salesforce, enabling cross-system data querying.
• Users can ask questions using natural language, and the AI will automatically integrate multi-source information to generate precise answers with cited sources. This aims to break down corporate "knowledge silos" and enhance decision-making efficiency. 💡
• The feature is powered by an enhanced GPT-5 model optimized specifically for multi-source retrieval. OpenAI also pledges that all enterprise data will only be used for immediate queries and never for model training, ensuring the security of business secrets. 🔒

Related Links:

• ChatGPT's 'Enterprise Brain' Officially Online! OpenAI Launches Company Knowledge Feature, Seamlessly Integrating 12 Major Platforms like Slack and Drive, Skyrocketing Decision-Making Efficiency.

OpenAI 收购 Workflow 原班人马，强化桌面端 AI 自动化

简报：

• OpenAI 宣布收购了开发 AI 自动化工具 Sky 的公司 Software Applications, Inc.。
• Sky 的创始团队曾开发了知名应用 Workflow，该应用后来被苹果公司收购，成为 iOS 和 macOS 系统中的“快捷指令”功能。
• Sky 是一款尚未发布的 macOS 软件，旨在作为用户的个人助手，通过观察屏幕内容并自动执行写作、编程等操作。
• 收购完成后，Sky 团队将全部加入 OpenAI，致力于将 macOS 的深度集成能力与产品工艺引入 ChatGPT。
• 此次交易的财务细节未公开，Sky 此前已从投资者处筹集了 650 万美元资金，投资者包括 OpenAI CEO Sam Altman。

相关链接：

• OpenAI 收购苹果 “快捷指令” 原团队新创公司

OpenAI Acquires Original Workflow Team, Enhances Desktop AI Automation

Brief:

• OpenAI announced the acquisition of Software Applications, Inc., the company behind the AI automation tool Sky.
• Sky's founding team previously developed the well-known app Workflow, which was later acquired by Apple and became the 'Shortcuts' feature in iOS and macOS. 🚀
• Sky is an unreleased macOS software designed to act as a personal assistant for users, observing screen content and automating tasks like writing and coding. 🤖
• Following the acquisition, the entire Sky team will join OpenAI, focusing on integrating macOS's deep integration capabilities and product craftsmanship into ChatGPT.
• The financial details of the transaction were not disclosed. Sky had previously raised $6.5 million from investors, including OpenAI CEO Sam Altman.

Related Links:

• OpenAI Acquires Startup from Apple's 'Shortcuts' Original Team 🔗

AI赋能精准医疗：DeepGEM大模型投入临床加速癌症基因解码

简报：金域医学联合腾讯和广州医科大学附属第一医院，已将其自主研发的癌症病理基因大模型DeepGEM投入临床应用，该模型通过融合组织切片图像与基因测序等多模态数据，能精准预测关键致癌突变，辅助医生为患者制定个性化治疗方案。

相关链接：

• DeepGEM大模型落地！金域医学联手腾讯、广医一院，用AI精准“读取”癌症基因密码

AI-Powered Precision Medicine: DeepGEM Large Model Enters Clinical Use, Accelerating Cancer Gene Decoding 🚀

Brief: KingMed Diagnostics, in collaboration with Tencent and The First Affiliated Hospital of Guangzhou Medical University, has successfully deployed its independently developed large AI model for cancer pathology and genomics, DeepGEM, into clinical application. This innovative model integrates multi-modal data, including tissue section images and gene sequencing, to accurately predict key oncogenic mutations. This capability significantly assists doctors in formulating personalized treatment plans for patients. 🧬🔬

Related Link:

• DeepGEM Large Model Launched! KingMed Diagnostics, Tencent, and GMU First Affiliated Hospital Partner to Precisely "Read" Cancer Gene Codes with AI

Gemini加持谷歌地球，灾害预警从数周缩至几分钟

简报：

• Google于2025年10月26日发布集成Gemini大模型的Google Earth AI，通过新增的“地理空间推理”能力，将灾害风险分析的耗时从数周缩短至几分钟。
• 用户可使用自然语言提问，系统能结合天气、人口、地形等多维数据，一键预测洪水淹没范围、野火蔓延路径及有害藻华爆发等灾害。
• 该系统由Imagery（图像）、Population（人口）和Environment（环境）三大专业基础模型支撑，能够主动识别地表变化、追踪人群流动和监测环境健康。
• 其核心模型已部署至Google Cloud平台，面向专业用户提供订阅服务，起步价为每月75美元，旨在将应急响应时间缩短25%以上。

相关链接：

• Google Earth AI“地球大脑”上线！Gemini一键预测洪水、藻华、野火，全球灾害预警从数周缩短至几分钟

Gemini Empowers Google Earth: Disaster Warnings Cut from Weeks to Minutes

Brief:

• Google launched Google Earth AI on October 26, 2025, integrating the Gemini large language model. This new platform, with its "geospatial reasoning" capability, slashes disaster risk analysis time from weeks to mere minutes. 🌍
• Users can pose questions in natural language, and the system leverages multi-dimensional data including weather, population, and terrain to instantly predict disasters such as flood inundation zones, wildfire spread paths, and harmful algal bloom outbreaks with a single click. 🔥
• The system is underpinned by three specialized foundational models—Imagery, Population, and Environment—which actively identify surface changes, track population movements, and monitor environmental health.
• Its core models are already deployed on the Google Cloud platform, offering subscription services to professional users starting at $75 per month, with the goal of reducing emergency response times by over 25%. 🚀

Related Links:

• Google Earth AI "Earth Brain" Launched! Gemini Predicts Floods, Algae Blooms, Wildfires with One Click, Global Disaster Warnings Reduced from Weeks to Minutes

蚂蚁百灵开源新模型，长序列推理吞吐量提升近三倍

简报：

• 蚂蚁百灵大模型团队开源了新一代高效推理模型 Ring-mini-sparse-2.0-exp，该模型基于 Ling2.0 架构。
• 模型专为长序列解码优化，采用了创新的稀疏注意力机制，并结合了高稀疏比的 Mixture of Expert（MoE）结构。
• 得益于架构与推理框架的协同优化，新模型在处理长序列时的吞吐量相比其前身 Ring-mini-2.0 提升了近三倍。
• 该模型在多项高难度推理基准测试中保持了领先（SOTA）性能，为开源社区提供了新的轻量化解决方案。

相关链接：

• 蚂蚁百灵团队发布新一代高效推理模型 Ring-mini-sparse-2.0-exp
• Ring-V2 GitHub 开源地址

Ant Bailian Open-Sources New Model, Nearly Tripling Long-Sequence Inference Throughput 🚀✨

Briefing:

• Ant Bailian's large model team has open-sourced a new generation of efficient inference model, Ring-mini-sparse-2.0-exp, which is based on the Ling2.0 architecture.
• The model is specially optimized for long-sequence decoding, adopting an innovative sparse attention mechanism and integrating a high-sparsity Mixture of Experts (MoE) structure.
• Benefiting from the collaborative optimization of its architecture and inference framework, the new model's throughput when processing long sequences has nearly tripled compared to its predecessor, Ring-mini-2.0.
• The model maintains State-of-the-Art (SOTA) performance in multiple challenging inference benchmarks, offering a new lightweight solution to the open-source community. 🔥

Related Links:

• Ant Bailian Team Releases New Generation of Efficient Inference Model Ring-mini-sparse-2.0-exp
• Ring-V2 GitHub Open-Source Address

美团发布AI视频模型，可生成5分钟连贯视频

简报：

• 美团公司推出了一款名为 LongCat-Video 的全新视频生成模型，其技术特点是能够原生支持生成长达5分钟的连贯视频。

相关链接：

• 美团推出 LongCat-Video 视频生成模型，原生支持5分钟级连贯输出

Meituan Unveils AI Video Model Capable of Generating 5-Minute Coherent Videos 🚀

Brief:

• Meituan Inc. has launched a new video generation model called LongCat-Video, which natively supports the generation of coherent videos up to 5 minutes in length. ✨

Related Link:

• Meituan Launches LongCat-Video Video Generation Model, Natively Supporting 5-Minute Coherent Output 🎞️